Network Working Group G. Bernstein, L. Ong
Internet Draft Ciena
Expiration Date: May 2002 B. Rajagopalan
Document: draft-ietf-ipo-optical-inter-domain-00.txt D. Pendarakis
Tellium
Angela Chiu
Celion
Frank Hujber
---
John Strand
AT&T
V. Sharma
Metanoia
Sudheer Dharanikota
Nayna Networks
Dean Cheng
Polaris
Rauf Izmailov
NEC
November 2001
Optical Inter Domain Routing Considerations
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026 [1].
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts. Internet-Drafts are draft documents valid for a maximum of
six months and may be updated, replaced, or obsoleted by other
documents at any time. It is inappropriate to use Internet- Drafts
as reference material or to cite them other than as "work in
progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
Abstract
This draft investigates the requirements for general inter-domain
and inter-area routing in optical networks and reviews the
applicability of existing route protocols in various optical routing
applications.
Table of Contents:
Bernstein, G. [Page 1]
draft-ipo-optical-inter-domain-00.txt November 2001
1 Introduction 3
1.1 Specification of Requirements 4
1.2 Abbreviations 4
2 Background 4
2.1 Basic Concept of Domains and Network Partitioning 4
2.2 Major Differences between Optical and IP datagram Routing 6
2.3 Diversity in Optical Routing 7
2.3.1 Generalizing Link Diversity 8
2.3.2 Generalizing Node Diversity 9
2.4 Routing Information Categorization 9
2.4.1 Link and Topology Related Information 10
2.4.2 Domain and Node Related Information 10
3 Applications of Inter Domain Optical Routing 11
3.1 Intra Carrier Applications of Optical Inter Domain Routing 11
3.1.1 Intra-Carrier Scalability 12
3.1.2 Intra-Carrier Inter-vendor 13
3.1.3 Inter-Layer Partitioning 15
3.1.4 Interaction with IP Layer Routing 17
3.1.5 Inter-Business Unit 17
3.2 Inter-Carrier Inter-Domain Optical Routing 20
3.3 Multi-Domain Connection Control 22
4 Multiple Layers of Routing 23
4.1 Layers in Transport Networks 23
4.2 Layer Integration 23
4.3 Interaction with IP Layer Routing 25
5 Existing Routing Protocol Applicability 25
5.1 OSPF 26
5.1.1 Terminology 26
5.1.2 Neighbor/Adjacency Discovery 27
5.1.3 Addressing & Reachability 29
5.1.4 Topology Discovery & Dissemination 29
5.1.5 Resources 30
5.1.6 General Protocol Properties 30
5.1.6.1 System Overhead 30
5.1.6.2 Network Resource Overhead 31
5.1.6.3 Reliability 31
5.1.7 Scaling Capability 32
5.1.8 Interworking Capability 33
5.1.9 References 33
5.2 IS-IS and Integrated IS-IS 33
5.2.1 Terminology 33
5.2.2 Neighbor/Adjacency Discovery 34
5.2.3 Addressing & Reachability 34
5.2.3.1 CLNP/NSAP addressing and routing 34
5.2.3.2 IP addressing and routing 35
5.2.4 Topology Discovery & Dissemination 36
Bernstein, G. [Page 2]
draft-ipo-optical-inter-domain-00.txt November 2001
5.2.5 Resources 36
5.2.6 Interface types and network Medium Support 37
5.2.7 General Protocol Properties 37
5.2.7.1 System Overhead 37
5.2.7.2 Network Resource Overhead 37
5.2.7.3 Reliability 39
5.2.8 Scaling Capability 39
5.2.9 Interworking Capability 39
5.2.10 Inter Domain Routing Protocols 40
5.2.11 References 40
5.3 BGP 40
5.3.1 Terminology 40
5.3.2 Neighbor/Adjacency Discovery 40
5.3.3 Addressing & Reachability 41
5.3.4 Topology Discovery & Dissemination 42
5.3.5 Scaling Capabilities 42
5.3.6 Interworking Capability 42
5.3.7 References 42
5.4 PNNI routing 42
5.4.1 Terminology 43
5.4.2 Neighbor/Adjacency Discovery 44
5.4.3 Addressing & Reachability 46
5.4.4 Topology Discovery & Dissemination 47
5.4.5 Resources 48
5.4.6 General Protocol Properties 49
5.4.6.1 System Overhead 49
5.4.6.2 Network Resource Overhead 49
5.4.6.3 Reliability 50
5.4.7 Scaling Capability 50
5.4.8 References 52
6 Conclusion 52
7 Security Considerations 52
7.1.1 53
8 References 53
9 Acknowledgments 54
10 Author's Addresses 55
1 Introduction
Multi Protocol Label Switching (MPLS) has received much attention
recently for use as a control plane for non-packet switched
technologies. In particular, optical technologies have a need to
upgrade their control plane as reviewed in reference [2]. Many
different optical switching and multiplexing technologies exist and
more are sure to come. For the purposes of this draft we only
consider non-packet (i.e. circuit switching) forms of optical
switching.
Bernstein, G. [Page 3]
draft-ipo-optical-inter-domain-00.txt November 2001
As the requirements for and extensions to interior gateway protocols
such as OSPF and IS-IS have begun to be investigated in the single
area case, e.g., reference [3], we consider the requirements that
optical networking and switching impose in the inter-domain case.
By inter-domain in this draft we consider inter-area, inter-layer,
and inter-vendor partitioning of routing and possibly other
possibilities for partitioning routing in addition to administrative
inter-domain (inter-carrier) partitioning. Comparisons of these
requirements to existing functionality in BGP, OSPF, IS-IS and ATM's
PNNI routing protocol are made.
In particular, optical routing needs to provide for path diversity,
switching capabilities, transport capabilities and impairments, and
bandwidth/resource status reporting.
To add to the concreteness of these considerations we try to
illustrate them with one or more specific examples from a particular
optical networking layer or technology. This is not to reduce the
generality of the requirement but to facilitate the understanding of
the requirement or concept.
1.1 Specification of Requirements
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
NOT","SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in
this document are to be interpreted as described in RFC 2119.
1.2 Abbreviations
LSP Label Switched Path (MPLS terminology)
LSR Label Switched Router (MPLS terminology)
MPLS Multiprotocol Label Switching
SDH Synchronous Digital Hierarchy (ITU standard)
SONET Synchronous Optical NETwork (ANSI standard)
STM(-N) Synchronous Transport Module (-N)
STS(-N) Synchronous Transport Signal-Level N (SONET)
TU-n Tributary Unit-n (SDH)
TUG(-n) Tributary Unit Group (-n) (SDH)
VC-n Virtual Container-n (SDH)
VTn Virtual Tributary-n (SONET)
2 Background
2.1 Basic Concept of Domains and Network Partitioning
In this draft we use the term domain in a very general sense, i.e.,
beyond the BGP interpretation of Administrative Domain. In this
draft we will consider domains as the result of partitioning of a
network into subnetworks, as shown in the network of Figure 1. A
network may be partitioned for a variety of reasons, such as:
* Administrative boundaries
* Scalability of routing or signaling
* Isolation of partitions for security or reliability reasons
Bernstein, G. [Page 4]
draft-ipo-optical-inter-domain-00.txt November 2001
* Technology differences in the systems in different domains
The Inter-Domain interface is likely to have different
characteristics than the Intra-Domain interface as the domain
boundary exists for the purpose of hiding some aspect within the
domain from the outside world.
Examples of the use of Domains include BGP Autonomous Systems (AS)
and OSPF Areas. An Administrative Domain AS is a subnetwork under
the control of single administration as viewed from outside of the
domain. In BGP case this is used to denote an interface between two
separate carriers or ISP. In the terminology used here we would call
this an inter-carrier domain interface. An OSPF Area, on the other
hand, is a subnetwork within an administrative domain (OSPF
specific) and results from partitioning of domains within a single
carrier, i.e., the interface to an Area is an intra-carrier domain
interface.
/------------------------------------\
/ \
/ /-\ \
| Domain |NE2|<--Internal Node |
| B /\-/\ |
| / \ |
| /-\/ \/-\ |
| |NE1|-------|NE3| /
\+------\-/ \-/ /
/\ | | /
/ \------+-----------+-----------------/
/ | |<--- Inter-Domain Links
/ /------+-----------+-----------------\
| / | | \
| / | /-\ | /-\ \
| | Domain | |NE2|---+---------|NE3| |
| | A | /\-/\ | ------/\-/ |
| | | / \ | / | |
| | /-\/ \/-\/ | |
| | |NE1|-------|NE5|---- | /
| \ -- \-/ \-/ \ /-\ /
| \ / | | --|NE4| /
| \/ | | \-/ /
| /\-----+-----------+----------------/
Border Nodes | |<--- Inter-Domain Links
\ |<----------+---- (External Links)
/\------+-----------+-----------------\
/ \ | | \
/ \ /-\ /-\ |
| --|NE1|-----|NE2| |
| \-/ \-/ |
| Domain | |<--- Intra-Domain |
Bernstein, G. [Page 5]
draft-ipo-optical-inter-domain-00.txt November 2001
| C | | Links (Internal|
| /-\ /-\ Links) |
| |NE3|-----|NE3| |
| \-/ \-/ |
\ /
\------------------------------------/
Figure 1. Partitioning a network into domains.
The domain concept as used here is orthogonal to the transport
network concept of layering. When the term layer is used with
respect to the transport network we are not referring to the 7-
layer OSI model which includes application, presentation, etc.,
layers. With regard to this model all the optical transport layers
would lie at the "physical layer". In the transport network, layers
are used for multiplexing, performance monitoring and fault
management purposes. Layers tend to be very technology specific. At
some point an optical routing protocol must include information
particular to the technology layer for which it is being used to
acquire/disseminate topology and resource status information. For
more information on layering and domain concepts see reference
[G.805].
2.2 Major Differences between Optical and IP datagram Routing
Let us first review the major difference between routing for optical
(circuit switched networks) and IP datagram networks. In IP
datagram networks packet forwarding is done on a hop-by-hop basis
(no connection established ahead of time). While circuit switched
optical networks end to end connections must be explicitly
established based on network topology and resource status
information. This topology and resource status information can be
obtained via routing protocols. Note that the routing protocols in
the circuit switch case are not involved with data (or bit)
forwarding, i.e., they are not "service impacting", while in the IP
datagram case the routing protocols are explicitly involved with
data plane forwarding decisions and hence are very much service
impacting.
This does not imply routing is unimportant in the optical case, only
that its service impacting effect is secondary. For example,
topology and resource status inaccuracies will affect whether a new
connection can be established (or a restoration connection can be
established) but will not (and should not) cause an existing
connection to be torn down.
This tends to lead to a slightly different view towards
incorporating new information fields (objects, LSA, etc.) into
optical routing protocols versus IP routing protocols. In the
optical circuit case, any information that can potentially aid in
route computations or be used in service differentiation may be
Bernstein, G. [Page 6]
draft-ipo-optical-inter-domain-00.txt November 2001
incorporated into the route protocol, as either a standard element
or a vendor specific extension. Whether a route computation
algorithm uses this information and whether two route computation
algorithms use this information in the same way doesnt matter since
the optical connections are explicitly routed (although perhaps
loosely). The optical route computation problem is really a
constraint-based routing problem. The basic route calculation is an
atomic service that occurs, for a given connection, in a single
network element. (In the case of loose explicit routing some details
may be filled in by other NEs.) This means that, even in a
heterogeneous optical network, NEs from different vendors need not
use the same algorithm.
Another difference - clear, hard blocking prevails in the optical
world while some level of overloading is ok in the IP world, i.e.,
statistical multiplexing is not available with optical circuits.
This also manifests itself in the commitment of the protection (or
restoration) bandwidth. In a packet-based network although the
protection path can be setup prior to any fault, the resources along
the protection path are not used until the failure occurs. In
circuit-based networks a protection path generally implies a
committed resource. Such a basic difference restricts the direct
applicability of some of the traffic engineering mechanisms used in
a packet-based network to a circuit-based network.
2.3 Diversity in Optical Routing
There are two basic demands that drive the need to discover diverse
routes for establishing optical paths:
1. Reliability/Robustness
2. Bandwidth capacity.
Many times multiple optical connections are set up between the same
end points. An important constraint on these connections is that
they must be diversely routed in some way [4]. In particular they
could be routed over paths that are link diverse, i.e., two
connections do not share any common link. Or the more stringent
constraint that the two paths should be node diverse, i.e., the two
paths do not traverse any common node.
Additionally, insufficient bandwidth may exist to set up all the
desired connection across the same path (set of links) and hence we
need to know about alternative (diverse) ways of reaching the
destination that may still have unused capacity.
"Diversity" is a relationship between lightpaths. Two lightpaths are
said to be diverse if they have no single point of failure. In
traditional telephony the dominant transport failure mode is a
failure in the interoffice plant, such as a fiber cut inflicted by a
backhoe.
Bernstein, G. [Page 7]
draft-ipo-optical-inter-domain-00.txt November 2001
Data network operators have relied on their private line providers
to ensure diversity and so IP routing protocols have not had to deal
directly with the problem. GMPLS makes the complexities handled by
the private line provisioning process, including diversity, part of
the common control plane and so visible to all.
Diversity is discussed in the IPO WG document [5]. A key associated
concept, "Shared Risk Link Groups", is discussed in a number of
other IETF (refs) and OIF (refs) documents. Some implications for
routing that are drawn in [5] are:
. Dealing with diversity is an unavoidable requirement for
routing in the optical layer. It requires dealing with
constraints in the routing process but most importantly
requires additional state information the SRLG relationships
and also the routings of any existing circuits from which the
new circuit is to be diverse to be available to the routing
process.
. At present SRLG information cannot be self-discovered. Indeed,
in a large network it is very difficult to maintain accurate
SRLG information. The problem becomes particularly daunting
whenever multiple administrative domains are involved, for
instance after the acquisition of one network by another,
because there normally is a likelihood that there are diversity
violations between the domains. It is very unlikely that
diversity relationships between carriers will be known any time
in the near future.
- Considerable variation in what different customers will mean by
acceptable diversity should be anticipated. Consequently we suggest
that an SRLG should be defined as follows: (i) It is a relationship
between two or more links, and (ii) it is characterized by two
parameters, the type of compromise (shared conduit, shared ROW,
shared optical ring, etc.) and the extent of the compromise (e.g.,
the number of miles over which the compromise persisted). This will
allow the SRLGs appropriate to a particular routing request to be
easily identified.
2.3.1 Generalizing Link Diversity
Optical networks may posses a number of hierarchical signaling
layers. For example two routers interconnected across an optical
network may communicate with IP packets encapsulated within an STS-
48c SONET path layer signal. Within the optical network this STS-
48c signal may be multiplexed at the SONET line layer into an OC-192
line layer signal. In addition this OC-192 may be wavelength
division multiplexed onto a fiber with other OC-192 signals at
different wavelengths (lambdas). These WDM signals can then be
either lambda switched, wave band switched or fiber switched. Hence
when we talk about diversity we need to specify the layer to which
we are referring. In the previous example we can talk about
Bernstein, G. [Page 8]
draft-ipo-optical-inter-domain-00.txt November 2001
diversity with respect to the SONET line layer, wave bands, and/or
optical fibers. A similar situation arises when we consider the
definition of node diversity. For example are we talking with
respect to a SONET path layer switch or an optical switch or
multiplexer?
The Shared Risk Link Group concept in reference [6] generalizes
the notion of link diversity (general list of numbers). First it's
useful with respect to major outages (cable cuts, natural disasters)
to have a few more types of diversity defined:
1. Cable (conduit) diversity (allows us to know which fibers
are in the same cable (conduit). This helps avoid sending
signals over routes that are most vulnerable to "ordinary"
cable cuts (technically known as backhoe fades).
2. Right of Way (ROW) diversity. This helps avoid sending
signals over routes that are subject to larger scale
disasters such as ship anchor drags, train derailments, etc.
3. Geographic Route diversity. This type of diversity can help
one avoid sending signals over routes that are subject to
various larger scale disasters such as earthquakes, floods,
tornadoes, hurricanes, etc. A route could be approximately
described by a piecewise set of latitude/longitude or UTM
coordinate pairs.
We also have a form of link abstraction/summarization via the link
bundling concept [7].
2.3.2 Generalizing Node Diversity
The concept of a node abstraction associated with GMPLS appears in
reference [11] where it is used to generalize the concept of an
explicitly routed path. In this case an abstract node can be a set
of IP addresses or an AS number. From the point of view of node
diverse routing specific concepts of interest include:
1. Nodes, i.e., individual switching elements.
2. Switching centers, i.e., a central office or exchange site.
3. Cities, or towns that contain more that one switching center.
4. Metro areas, or counties
5. States,
6. Countries, or
7. Geographic Regions
2.4 Routing Information Categorization
Different applications of inter-domain optical routing call for
different types of information to be shared or hidden between
domains. In the following we decompose the information that can be
transferred via a routing protocol broadly into link/topology
information and node/domain information. We further subdivide these
Bernstein, G. [Page 9]
draft-ipo-optical-inter-domain-00.txt November 2001
categories and will use this taxonomy of routing information when
discussing the routing applications.
2.4.1 Link and Topology Related Information
-Internal topology- information is information concerning the nodes
and links and their connectivity within a domain. This type of
information is traditionally shared within a domain via an intra-
domain (interior gateway) routing protocol such as OSPF or IS-IS.
For example the existence of nodes that only have links to other
nodes within the domain, i.e., do not have links to other domains,
would be strictly internal topology information. These nodes are
known as internal nodes, while nodes with links to other domains are
known as border nodes. Also included in this information is
link/port property information such as whether the link is protected
and what type of protection is being used, e.g., linear 1+1, linear
1:N, or some type of ring such as a 4F-BLSR [cite T1 document].
-Internal Resource- Information is concerned with the bandwidth
available on links within a domain and possibly other resource
related information. This information plays an important role in
path selection within a domain.
-Inter-Domain Topology- Information is concerned with how the
domains are interconnected. This information can be key in inter-
domain path selection, for example, in determining diverse routes.
For the network in Figure 1 this information would let us know that
domain A has two distinct links to domain B, domain A has two
distinct links to domain C, but that domains B and C are not
directly connected via any links.
-Inter-Domain Resource- Information is concerned with the available
bandwidth on inter-domain links. This information is important for
inter-domain path selection and inter-domain traffic engineering
purposes. For example in Figure 1 this information would give us
some kind of bandwidth or capacity measure on the links between
domain A and domain B, and the links between domains A and C. The
exact nature of this information may be application/context
dependent.
2.4.2 Domain and Node Related Information
-Reachability- information tells us what addresses are directly
reachable via a particular domain. These systems can be end systems
(clients) to the network or nodes within the network depending upon
the application/context. Suppose in domain B of Figure 1 each of
the network elements, NE1-NE3, have subtending end systems, and that
NE1-NE3 do not represent a valid final destination for a path. Under
Bernstein, G. [Page 10]
draft-ipo-optical-inter-domain-00.txt November 2001
this assumption the collection of the addresses of all these
subtending end systems would form the reachability information for
domain B.
-Subnetwork Capability- information is concerned with the
capabilities or features offered by the domain as a whole. This
information is used in some applications where sharing the internal
topology and resource information is inappropriate. This information
can include: (a) Switching capabilities, (b) Protection
capabilities, (c) Some kind of overall available capacity measure,
(d) Reliability measures.
Examples:
1. For example, in the SONET realm, one subnetwork may switch down
to an STS-3c granularity while another switches down to an STS-
1 granularity. Understanding what types of signals within a
SDH/SONET multiplex structure can be switched by a subnetwork
is important. Similar examples of granularity in switching
apply to the waveband case.
2. Some networking technologies, particularly SONET/SDH, provide a
wide range of standardized protection technologies. But not all
domains will offer all protection options. For example, a 2/4-
F BLSR based subnetwork could offer extra data traffic, ring
protected traffic and non-preemptible unprotected traffic,
(NUT)[8], while a mesh network might offer shared SONET line
layer linear protection and some form of mesh protection.
3. Some domains may be in locations that have lower incidences of
link failure. Such information could be helpful in computing
routes to statistically "share the pain".
-End System Capabilities- information: While properties of the
subnetwork are very important when trying to decide which domain to
use to access a system (in the case of multi-homing), end systems
also posses a wide variety of capabilities. Throwing end system
capabilities such as a system's ability to support SONET/SDH virtual
concatenation for distribution into a routing protocol may not be
that advantageous since it somewhat counters the ability to
summarize reachability information. Detailed end-system information
may alternatively be obtained via a directory service or some type
of direct query between the end systems.
3 Applications of Inter Domain Optical Routing
3.1 Intra Carrier Applications of Optical Inter Domain Routing
Intra Carrier inter domain routing refers to a situation where the
network that is to be partitioned into areas is under the control of
one administrative entity. The main reasons for this partitioning in
optical networks stem from scalability, inter-vendor
Bernstein, G. [Page 11]
draft-ipo-optical-inter-domain-00.txt November 2001
interoperability, legacy equipment interoperability, and inter-layer
partitioning.
3.1.1 Intra-Carrier Scalability
As networks grow it is useful to partition a carriers network into
separate optical routing domains which share limited or summarized
information amongst each other. This reduces the overhead of
information exchange across the network as a whole, and reduces the
convergence time of routing protocols within a particular area.
Hence we see in the inter-carrier scalability application that we
will hide or summarize internal topology and resource information,
while completely sharing inter-domain topology and resources
information so that diverse paths can still be calculated. Note that
general domain capabilities/capacity as well as reachability
information would tend to be shared as completely as possible. For
example the network shown in Figure 1 can be approximately
represented as shown in Figure 2. This summarized network topology
only has 4 links whose state need to be advertised in a routing
protocol versus 17 links in the original network. Note that we may
also advertise the capabilities of the three domains in Figure 2 as
opposed to the 12 nodes of Figure 1. In Note that this partitioning
into domains can recurse, i.e., we can have multiple levels of
routing hierarchy to permit larger and larger networks. Such was
the motivation behind the extensive hierarchical routing capability
within ATM's PNNI routing protocol. The trade off to partitioning
into domains for scalability is that less information is available
for use in the route selection process which can lead to inefficient
utilization of network resources. On the other hand frequently this
partitioning occurs on somewhat "natural" boundaries and as such the
potential inefficiencies can be minimized.
-------- ------
/- -\ /- -\ -----
// NE 3 \mmmmmm // \\ / \
/ Port 2 \ mmmmm NE 3 NE 5 \ /NE 3 \
/ \ / Port 4 Port 6 \ mmmmmPort 17 \
| | | mmmmm | |
| | | | | |
| | | Domain A | | Domain C |
| Domain B | | | | |
| | | NE 2 NE 1 | | |
| | |Port 5 Port 2 mmmmmmmmmNE 1 |
| mmm / \Port 21 /
\ NE 1 /mmmm \ / \ /
\ Port 7 mm \\ // \ /
\\ // \- -/ -----
\- -/ ------
Bernstein, G. [Page 12]
draft-ipo-optical-inter-domain-00.txt November 2001
--------
Figure 2. Summarized topology for the network of Figure 1.
Also in Figure 2 we show the end points of the links between being
identified by the triple of (domain, NE address, NE port number).
This information is available via the discovery process. Though not
strictly necessary including the identification of border nodes in a
domain, allowing other nodes to understand whether these links
terminate on the same or different nodes is valuable in setting up
diverse inter-domain paths.
In current intra-domain IP routing protocols, such as OSPF's, a
multiple area capability provides for intra-carrier scalability.
However, this is currently done by sharing reachability information
and using a vector distance method to obtain routes. This does not
discover and propagate inter-domain topology information and hence
insufficient information is available for diverse route
calculations.
When the topology within the domain is approximated or hidden then
signaling and call processing at the domain border will receive an
approximated (loose) route and the border node or signaling entity
must then translate this to a precise route through the domain.
Hence there is some linkage between multi-domain connection control
and inter-area/inter-domain routing.
3.1.2 Intra-Carrier Inter-vendor
An important application of intra-carrier optical routing is the
intra-carrier inter-vendor scenario. From a carriers perspective,
the use of domains provides a clean way to isolate clouds of
equipment belonging to different vendors, while at the same time
allowing for interoperability between the vendors.
An advantage of this method is that it allows the vendors complete
freedom to use any combination of routing protocols or traditional
management-based methods to propagate topology and resources
internal to their domains. In other words, the routing entity in
each domain could obtain this information either by participating in
a routing protocol like OSPF, or by querying each NE via an EMS, or
by simply having the required information manually configured into
it.
Note that the routing entity shown in each vendors cloud in Figure
3 is in reality an abstract representation of the routing
intelligence within the vendor cloud. This intelligence may either
be implemented in a distributed way, by having a routing protocol
running at each NE, or in a centralized way, through the use of an
intelligent, centralized routing entity that communicates with the
individual NEs (either via a protocol or by querying individual
Bernstein, G. [Page 13]
draft-ipo-optical-inter-domain-00.txt November 2001
elements) to retrieve connectivity and resource information that it
uses to build a complete topology and resource map of the domain.
Therefore, vendors may use different protocols as the primary option
between their own devices, adding specialized features or optimizing
their performance based on their choice of protocol.
/------------------------------------\
/ /-\ \
/ Domain B |NE3| +-------+ \
| (Vendor 2) /\-/\ |Routing| |
| / \ |Entity | |
| /-\/ \/-\ +-------+ |
\ |NE1|-------|NE2| @ /
\ \-/: \-/ @ /
\------+-:---------+---------@-------/
Neighbor discovery | : | @ Exchange of routing
Between NEs in the | : | @ information between
same domain. ----> | : | @ the domains routing
| : | @ entities.
/------+-:---------+---------@-------\
/ | : | @ \
/ /-\ /-\ +-------+ \
| |NE1|-----|NE2| |Routing| |
| \-/\ /\-/ |Entity | |
| \/-\/ +-------+ |
| Domain A |NE3| |
| (Vendor 1) \-/ /
\ /
\------------------------------------/
Figure 3. Intra-Carrier Inter-vendor routing domains
Even if it is a centralized entity, the routing entity could still
be run on a given NE in the vendor cloud. In other words, these
entities could be distributed, or centralized onto a single node, or
independent of any of the nodes. In ATM's PNNI protocol, for
example, this was centralized on a node elected as the "peer group
leader". In the inter-vendor case, it can be particularly
advantageous to centralize this so that the flow of information can
be monitored. A centralized routing entity could apply flooding and
summarization mechanisms as if it is a switching system. Since this
is optical rather than IP routing, signaling would be carried by a
control channel between the routing entity and the neighboring
system, rather than being carried over the data links. The functions
of the routing entity include: (a) direct reachability exchange
(that is, which NEs can be directly reached from this domain, (b)
verification of area connectedness (that is, understanding how the
two domains are interconnected), (c) area/domain topology (possibly
summarized) exchange and updates, and (d) topology updates
concerning other domains/areas.
Bernstein, G. [Page 14]
draft-ipo-optical-inter-domain-00.txt November 2001
When a carrier partitions its network for inter-vendor
interoperability as described above, it may still share information
about the internal topology of the domains in some standardized form
that has been agreed upon between the vendors. Although one option
is to force both vendors to adopt a new common protocol, another is
to require that only a minimum subset of reachability/topology
information be shared between the vendor clouds. The latter option
helps during fault situations, by providing fault isolation at the
domain boundaries. It prevents an outage in a domain composed of one
vendors equipment from causing a reaction in an adjacent domain
composed of another vendors equipment, thus preventing a situation
that would typically degenerate into a process known as finger
pointing between the two vendors.
The setup described above takes into account the three most
important sub-cases of inter-carrier inter-vendor partitioning:
a. The first is where both vendor domains run distributed
routing protocols. This is the most flexible case, and is the
situation when new equipment capable of running such protocols
is deployed.
b. The second is where the optical subnetworks or domains
(which includes a large number of existing installations) do
not run any internal routing protocol (because the NEs are not
capable of doing so), relying instead on EMS-based topology
discovery/resource management. In this case, interoperability
with other vendor clouds can be realized by having the routing
entity run as a separate software entity with access to the
appropriate information. These entities may exchange routing
proxy addresses through the neighbor discovery protocol, and
then exchange routing information (proxying for the entire
domain) with each other. The basic advantage here is that even
though the vendor specific element management system (EMS)
knows the topology of its subnetwork, it is far easier to get
an inter-domain routing protocol to share information than
trying to get the separate vendors management systems to
communicate.
c. The third is where one domain has a centralized routing
entity, while the other runs a distributed routing protocol.
Once again, the neighbor discovery process between the area
border NEs could be used to advertise the address of the
routing entity.
3.1.3 Inter-Layer Partitioning
In transport networks layering is a part of the multiplex and OA&M
structure of the signals, playing a role in multiplexing, monitoring
Bernstein, G. [Page 15]
draft-ipo-optical-inter-domain-00.txt November 2001
and general link management. Layering in the transport network is
defined in fairly abstract terms in [G.805] and the concepts are
applied to SDH in [G.803]. As explained in a recent ITU SG15
document (WD45 Q.14/15) not all the layers in the transport network
are of interest to the control plane, or to routing in particular.
Some layers may not contain active switching elements, however this
does not mean that information flow concerning a non-switching layer
is not valuable in routing. For example in [GB-WDM-SRLG] static WDM
layer information was used to set the SRLGs for SONET lines (i.e.,
information passed around by a link state protocol operating at the
SONET line layer). It should be noted that much of the information
available from non-switching layers relates to performance
monitoring and fault management.
In this situation the network is partitioned into sub-networks that
operate at different switching layers. One reason for doing this is
that not all the information from one layer is necessary or relevant
to another layer. For example, between transparent optical switches
and SDH/SONET path (VC) layer switches, the optical switches have no
direct use for the SONET layer information. In addition optical
networks may keep a lot more physical layer information (such as the
properties of every optical amplifier on a WDM span) that is of no
use to the SONET layer. One again this promotes scalability, but
also simplifies the implementation by reducing inter-layer
information transfer to that which is actually useful.
Now in network planning it is very useful to have a view of the
current higher layer traffic matrix [9] being satisfied and higher
layer traffic trend measurements over time. Although we can
somewhat see this in higher layer resource status changes over time,
this represents a link level view when we really desire the trend
(change in time) of the traffic matrices between sites. How this
information gets distributed is an open issue. Currently individual
nodes in a GMPLS network know only about connections that they
source or sink. Network planning is generally a longer time horizon
process than even traffic engineering hence it is an open question
as to whether this would ever be a useful function to incorporate
into a network element.
Now looking the other way is initially simpler, i.e., it is easier
to ask: what can a higher layer use for path selection/computation
from a lower layer. The first item that springs to mind is diversity
information. For example in setting up a SONET STS-1 path we can use
information from a WDM system concerning which SONET lines share
the same WDM fiber. This is information, however, is already
abstracted into routing protocols via SRLG concept. Other
Bernstein, G. [Page 16]
draft-ipo-optical-inter-domain-00.txt November 2001
information from a lower layer is of questionable value since it
tends to be technology specific and puts more and more burden on the
upper layer to be able to effectively understand and use this
information.
3.1.4 Interaction with IP Layer Routing
The applicability of IP-based routing protocols has, over the years,
been constantly expanded to increasingly more circuit-oriented
layers. The community began with pure datagram routing, gradually
expanded to cover virtual-circuit switched packet routing (for e.g.,
MPLS), and is finally looking at the application of routing
protocols to real circuit switching, e.g. the optical layer.
However, as pointed out earlier in this document, it is not clear
that the different layers should necessarily share the same instance
of the routing protocols. Indeed, there may be significant reasons
for not doing so and many carrier tend to partition there networks
along switching layer boundaries. For example, IP-layer reachability
information is not particularly useful for the optical layer, so it
seems an overkill to burden the optical equipment with storing and
distributing that information. (It is an extra expense on memory and
processing for information that the optical layer does not really
care about, so there is little incentive for a vendor to want to do
so.) Likewise, information on physical plant (fibers, conduits,
ducts) diversity, which is crucial at the optical transport layer,
is very unlikely to be used directly by the IP layer. So, it would
be quite wasteful of resources to burden the IP layer routing with
distributing and manipulating this information. Thus, the extent of
interaction or integration with IP layer routing (if any) requires
careful consideration.
3.1.5 Inter-Business Unit
A slightly different but interesting application of intra-carrier
optical routing occurs in the intra-carrier inter-business unit
scenario. This arises because a carrier often has multiple
administrative domains, with groups of administrative domains being
under the purview of independent BUs within the carrier.
Note that different BUs represent independent cost centers with
their own profit objectives and sales targets. As a result, while
the BUs can profitably share topology information and would like to
do so, they may not be so inclined to advertise the details of their
resource usage into domains belonging to other BUs. Since each BU
has its own revenue targets, advertising detailed resource
availability information to other, potentially competing, BUs can
have a negative impact on a BUs revenue generation. This is because
Bernstein, G. [Page 17]
draft-ipo-optical-inter-domain-00.txt November 2001
the knowledge of available resources in one BU may enable other BUs
within the carrier to requisition capacity from this BU. This would
force the BU in question to yield to their request, possibly at the
expense of selling capacity to more profitable, external, revenue
generating customers.
Thus, this scenario is likely to have an additional dimension of
information sharing, namely, policy-based information sharing, which
does not apply to the other cases that we have discussed so far.
Two examples where the inter-business unit scenario could become
important are in the case of metro-core-metro networks within a
given carrier, and in the case of regional networks within a
carrier.
In the metro-core-metro situation, such as the one depicted in
resource sharing between them.
/------------------------------------\
/ \
/ /-\ \
| Domain |NE2| |
Metro BU | B /\-/\<-- Metro Ring |
Network | / \ |
| /-\/ \/-\ |
| |NE1|-------|NE3| /
\ \-/ \-/ /
\ | | /
\------+-----------+-----------------/
| |
/------+-----------+-----------------\
/ | | \
/ | /-\ | /-\ \
| Domain | |NE2|---+---------|NE3| |
CORE BU | A | /\-/\ | ------/\-/ |
Network | | / \ | / | |
| /-\/ \/-\/ |<--Core |
| |NE1|-------|NE5|---- | Mesh /
\ \-/ \-/ \ /-\ /
\ | | --|NE4| /
\ | | \-/ /
\-----+-----------+----------------/
| |
/------+-----------+-----------------\
/ | | \
/ /-\ /-\ \
| |NE1|-----|NE2| |
Metro BU | Domain \-/ \-/ |
Network | C | |<--- Metro Ring |
| | | |
| /-\ /-\ |
Bernstein, G. [Page 18]
draft-ipo-optical-inter-domain-00.txt November 2001
| |NE3|-----|NE3| |
| \-/ \-/ |
\ /
\------------------------------------/
Figure 4, the metro network domains could be under the jurisdiction
of one BU, while the core network domains belong to a different BU.
In this case, for example, it is possible that the metro BU, armed
with resource availability information about the core BUs domains,
could requisition capacity from the core network when needed. This
may harm the core networks profit goals, because they may not be
able to charge an internal customer the same rates that they could
charge for the same capacity from an external customer, thus
motivating the need for selective, policy-based resource sharing
between them.
/------------------------------------\
/ \
/ /-\ \
| Domain |NE2| |
Metro BU | B /\-/\<-- Metro Ring |
Network | / \ |
| /-\/ \/-\ |
| |NE1|-------|NE3| /
\ \-/ \-/ /
\ | | /
\------+-----------+-----------------/
| |
/------+-----------+-----------------\
/ | | \
/ | /-\ | /-\ \
| Domain | |NE2|---+---------|NE3| |
CORE BU | A | /\-/\ | ------/\-/ |
Network | | / \ | / | |
| /-\/ \/-\/ |<--Core |
| |NE1|-------|NE5|---- | Mesh /
\ \-/ \-/ \ /-\ /
\ | | --|NE4| /
\ | | \-/ /
\-----+-----------+----------------/
| |
/------+-----------+-----------------\
/ | | \
/ /-\ /-\ \
| |NE1|-----|NE2| |
Metro BU | Domain \-/ \-/ |
Network | C | |<--- Metro Ring |
| | | |
| /-\ /-\ |
| |NE3|-----|NE3| |
| \-/ \-/ |
Bernstein, G. [Page 19]
draft-ipo-optical-inter-domain-00.txt November 2001
\ /
\------------------------------------/
Figure 4. Intra-carrier inter-business unit routing domains.
3.2 Inter-Carrier Inter-Domain Optical Routing
In this case we are talking about dealing with outside entities,
i.e., between service providers. There may be a range of levels of
trust here; for example there might be some level of trust between
two providers that have formed a marketing alliance or have some
other form of business relationship. In general, however, trust can
not be assumed. In this case, all the concerns of revealing too much
information about one's network come into play. However, not
revealing enough, say about diversity capabilities may also lead
customers elsewhere. Also there are some other security issues not
seen before. For example, in route distribution one carrier might
not be inclined to pass on routing information that could point the
way to competitive alternatives. This impacts the methods for route
updates, etc.
With the interest in bandwidth trading [10] we can also look at this
as an advertisement of network connectivity and capability with of
course any "warts" covered up. This would include reliance on other
carrier for fibers or lambdas. Also a fair amount of details such
as "unused capacity" would not be advertised since this maybe
financially sensitive information.
Private line pricing today is based primarily on the service itself
(bandwidth, end-points, etc.) and the holding time, and there is no
reason to expect that this will change. When multiple service
providers are involved the algorithm for dividing up the revenue
stream (which can be quite large even for a single connection) must
be explicit by connect time. This could be done off-line or could be
done at connect time. In either case, the entity or entities doing
the routing will need to take provider pricing structures into
account whenever there is a choice between providers that needs to
be made. The routing logic could do this explicitly if the prices
are captured in the advertised metrics or some other advertised
data; alternatively it could be done by some sort of policy control,
as it is today by BGP.
The essence of bandwidth trading is the existence of competing price
structures that are known to the entity deciding which competitor to
use. It is possible to create plausible bandwidth trading scenarios
involving the UNI, the NNI, or both. If the NNI is involved, these
price structures will need to be established across it. The
situation is further complicated by the fact that bandwidth trading
could be realized using any one of a number of business models, each
with its own information requirements. To give two examples: If an
auction model were used the buyer might repeatedly broadcast the
Bernstein, G. [Page 20]
draft-ipo-optical-inter-domain-00.txt November 2001
lowest bid received to date and solicit lower bids from the
competing providers. On the other hand, if there were a more formal
market the providers might post their asking prices in some public
fashion and a buyer would be matched by some third party with the
lowest offer.
In the inter-carrier case notions of hierarchy seem rather
sensitive, i.e., he who controls the summarization and advertisement
may have an undue advantage over competitors. In addition, a
"bandwidth aggregator" may want to advertise capabilities that he
has put together via deals with multiple carriers...
Notes: We can attempt to extend the SRLG concept to links between
ASs but we will need the two ASs to agree on the meaning and number
of the list of 32 bit integers that comprise the SRLG, i.e.,
previously the SRLG concept was one of AS scope. And this is also
where things get tricky since it may not be possible to distinguish
diverse routes based upon differing path vectors (i.e., AS number
traversal list). The reason for this is due the fact that many
carriers "fill out" their networks by renting either dark fiber or
"lambdas" from a WDM system and hence although the path vectors may
be AS diverse they may not even be fiber diverse.
Hence there is a need for sharing of diversity information or
constraints between ASs when setting up diverse connections across
multiple ASs. This gets us somewhat into a quandary over which
information needs to be public and how to coordinate its
distribution. In this sense geographic link information may be the
simplest and least contentious to get various players to disclose
and standardize.
Notes: (1) The real issue is consistency between the cloud/ASs
since in many cases they are sharing conduit, ROW, etc. Getting
this to happen could be very problematic. It would be preferable to
see a diversity option that doesnt require this. For example,
ensure that there is diversity within each cloud and then do
restoration separately within each cloud. (2) See the definition of
SRLG in the Carrier Requirements an equivalence class of links,
the extent of violation, and the level. (3) Flexibility in defining
the level of violation seems very desirable these historically
have drifted in time. There are many others eg, if the shared
resources are SPRING protected thats less of a problem than
otherwise.
Notes: Participation in the inter-domain network carries constraints
on the carriers. First, in order to participate, each provider
network MUST be willing to advertise the destinations that are
reachable through his network at each entry point and advertise the
formats available. Without providing such information, there is
little motivation to participate since it is unlikely that others
will be able to access services of which they are not aware. Second,
every participating carriers MUST agree to fairly include the
information made available by every other carrier so that each
Bernstein, G. [Page 21]
draft-ipo-optical-inter-domain-00.txt November 2001
carrier has an equal opportunity to provide services. There may be
specific exceptions, but the carrier claiming those exceptions MUST
advertise the exceptions themselves. In this manner, other carriers
that might otherwise be aware of distant services can be prompted to
seek those services manually. Note a combination of minimal
required information transferred with deferral to the originating
subnetwork along with some basic security mechanisms such as
integrity and non-repudiation may be useful in helping organizations
to "play nice".
3.3 Multi-Domain Connection Control
MPLS loose routing capability allows one to specify a route for an
optical connection in terms of a sequence of optical AS numbers.
This, for example, is handled via RSVP-TEs abstract node concept
[11]. Currently there is nothing in the GMPLS signaling
specification that differentiates between intra AS boundaries, i.e.,
between two neighbor optical LSRs in the same AS, and inter AS
boundaries, i.e. between two neighbor optical LSRs in different ASs.
Note that these same notions can apply to separate routing domains
within an AS. There may, however, be some useful reasons for
differentiating these two cases:
1. Separation of signaling domains,
2. Separation of protection domains.
While routing protocols (used for their topology information) in the
optical case are not "service impacting", signaling protocols most
certainly are. It is desirable to build some type of "wall" between
optical ASs so that faults in one that lead to "signaling storms" do
not get propagated to other ASs. Note that the same motivation
applies for isolating other kinds of clouds, like vendors specific
ones.
The natural situation where "signaling storms" would be most likely
to arise is during network restoration signaling, i.e., signaling to
recover connections during major network outages, e.g., natural
disasters etc. In this case it may be very advantageous to break up
general source reroute forms of restoration into per domain segments
or to start reroute at domain boundaries rather than all the way
back at the originating node. Note that this has the advantage of
reducing the need for globally consistent SRLGs. (See earlier SRLG
comment.) Such a capability requires some loose coordination between
the local, intermediate and global protection mechanisms [12]. This
is typically implemented via hold off timers, i.e., one layer of
protection will not attempt restoration until a more fundamental
(local) form has been given a chance to recover the connection [12].
In other words, prevention of restoration related signaling storms
may require the breaking up of a large network into multiple
signaling (and hence routing) domains. These domains could be within
the same AS.
Bernstein, G. [Page 22]
draft-ipo-optical-inter-domain-00.txt November 2001
4 Multiple Layers of Routing
4.1 Layers in Transport Networks
In transport networks layering is a part of the multiplex and OA&M
structure of the signals, playing a role in multiplexing, monitoring
and general link management. Layering in the transport network is
defined in fairly abstract terms in [G.805] and the concepts are
applied to SDH in [G.803]. As explained in a recent ITU SG15
document (WD45 Q.14/15) not all the layers in the transport network
are of interest to the control plane, or to routing in particular.
Some layers may not contain active switching elements, however this
does not mean that information flow concerning a non-switching layer
is not valuable in routing. For example in [GB-WDM-SRLG] static WDM
layer information was used to set the SRLGs for SONET lines (i.e.,
information passed around by a link state protocol operating at the
SONET line layer). It should be noted that much of the information
available from non-switching layers relates to performance
monitoring and fault management. Hence work in this area within
CCAMP should take into account this layered approach.
Note that this is distinct from the layer idea used in the 7-layer
OSI model or IP layer model. In the IP model, the term Layer means
that, for example, the Application Layer entity requests services
for delivering a message to an entity on another computer and it
contacts the Transport Layer service, which in turn contacts the
Internet Layer. Lower layers are successively contacted until an
end-to-end service is provided. A key concept is that the
Application Layer cannot (or rather should not) contact the Internet
Layer directly. In this model all the "layers" discussed in this
document would lie in the "physical layer" (from an IP perspective).
4.2 Layer Integration
As previously discussed, there are multiple layers of signals
included in what in the IP model one would call the Physical Layer.
One could separate the layers by creating sublayers in the Physical
Layer. For example, sublayers in the Physical Layer might be, top
to bottom: LOVCs, HOVCs, and Lambdas. If a system supports only one
of the three, then isolation of the sublayers is a given; it's
geographical. But there are systems which will support more than one
physical sublayer, therefore, it is necessary to establish whether
or not there is a need to isolate the sublayers in the same manner.
Or put another way is there a reason to "integrate" the sublayers
for the purposes of routing (topology dissemination).
If they are isolated, then there will be separate topological models
for each sublayer: one mesh for the LOVC, one for the HOVC, one for
the Lambda, and possibly others. The appropriate way to access a
sublayer is via the use of sublayer SAPs (service access points).
For example, in this way, one may find that use of Lambdas is more
Bernstein, G. [Page 23]
draft-ipo-optical-inter-domain-00.txt November 2001
efficient because each sublayer can assess the availability of
services at its own layer before searching for coarser-granularity
services. On the other hand, the control plane must accommodate
three separate routing protocols, or at least three separate
instances of the same routing protocol, all operating at both intra
and inter-domain level.
Section 4.4.2, herein, states "For transport across a SONET network,
the lower order signals must be multiplexed into a non-concatenated
higher order signal." Given that this is true, LOVCs are not routed
independently, but only as tributaries of HOVCs. In addition in the
SDH hierarchy there is a signal, VC3, that can be treated
(multiplexed) as either a LOVC or a HOVC. With this tight and
somewhat confused coupling of these layers it may beneficial to
sometimes combine them into the same route protocol instance.
Use of the terms LOVC and HOVC infers that all of the services to be
supported by inter-domain routing are those formally associated with
the terms in SONET and SDH standards. However, among the optical
systems emerging in todays market are rate and format independent
systems, which claim to offer services that do not rely on SONET/SDH
framing. Their intent is to support Ethernet, ATM, and OTN framing
without the need for electronics specifically targeted at the signal
of interest. The question arises whether or not to include these
"clear channel" services as a separate sublayer of the Physical
Layer.
The alternative to separate routing protocols per sublayer is the
original notion behind GMPLS routing and the forwarding adjaciency
concept [13]. Rather than separating the route protocols into
separate layers (or sublayers) with distinct topologies, each ONE
would advertise the services it can provide, along with its topology
information. For example, a ONE (optical network element) might
advertise that it carries a route to node A with STS-N service and
clear-channel lambda service and carries multiple routes to node B
with STS-N service. It might, alternatively, advertise its entire
network with summarized link capacity information for every included
link. Neighboring carriers would, implicitly, be allowed to
summarize that information for internal advertisement via its IGP.
Further consideration could be given to a query service, where a
carrier advertises the geographical area it serves without detailed
reachability or capacity information. A second carrier desiring
service could query the first carrier as to reachability for a
specific destination, and the first carrier would respond with
availability and capacity information.
Integrating multiple layers into the same routing protocol instance
leaves us fewer routing protocols to manage. The downside of this is
that more information must be exchanged via this routing protocol
and more network elements participate in this single instance of the
routing protocol which can lead to scalability concerns. If the
equipment working on the different sublayers comes from different
Bernstein, G. [Page 24]
draft-ipo-optical-inter-domain-00.txt November 2001
vendors there would be little incentive to integrate multiple layers
into the routing protocol for a single layer product.
Regardless of whether multiple layers are integrated into the same
routing protocol instance it can be very useful to share information
between layers as illustrated by the following examples:
o Drop side links between layers: Capabilities of the links
that are between the (client and server) layers need to be
propagated into the routing protocol.
o Summarize link capabilities: Summarizing the server layer
capabilities in the client layer will reduce the amount of
information required for multi-layer constraint based path
computation.
o Send only that are required: Sending only the capabilities
that are useful in the constraint path computation in the
client layer.
4.3 Interaction with IP Layer Routing
The applicability of IP-based routing protocols has, over the years,
been constantly expanded to increasingly more circuit-oriented
layers. The community began with pure datagram routing, gradually
expanded to cover virtual-circuit switched packet routing (for e.g.,
MPLS), and is finally looking at the application of routing
protocols to real circuit switching, e.g. the optical layer.
However, as pointed out earlier in this document, it is not clear
that the different layers should necessarily share the same instance
of the IP routing protocols. Indeed, there may be significant
reasons for not doing so. For example, IP-layer reachability
information is not particularly useful for the optical layer, so it
seems an overkill to burden the optical equipment with storing and
distributing that information. (It is an extra expense on memory and
processing for information that the optical layer does not really
care about, so there is little incentive for a vendor to want to do
so.) Likewise, information on physical plant (fibers, conduits,
ducts) diversity, which is crucial at the optical transport layer,
is very unlikely to be used directly by the IP layer. So, it would
be quite wasteful of resources to burden the IP layer routing with
distributing and manipulating this information.
Thus, the extent of interaction or integration with IP layer routing
(if any) requires careful consideration.
5 Existing Routing Protocol Applicability
Here we look at the applicability of OSPF,IS-IS PNNI and BGP to
various aspects of the general optical inter domain routing problem.
All protocols provide reachability information. The questions to be
investigated are how they deal with partitioning the network,
Bernstein, G. [Page 25]
draft-ipo-optical-inter-domain-00.txt November 2001
diverse routing, summarized/abstracted topology information sharing,
and suitability for the inter-domain environment.
5.1 OSPF
OSPF stands for Open Shortest Path First, an IP interior routing
protocol defined by the IETF as documented as the RFC2328. OSPF uses
the link state algorithm. OSPF was originally defined to construct
shortest path tree in IP networks to forward IP datagram in hop-by-
hop manner in IPv4 addressing space. The OSPF is capable of scaling
to a very large networks by partitioning networks into areas. Each
OSPF area is identified by a unique identifier called Area ID. The
default area is called backbone area with the Area ID as zero. A
router with interfaces with multiple OSPF areas is called Area
Border Router (ABR). The topology information including node, links
and addresses belonging to a single area are advertised throughout
that area but not beyond. Only reachability of address and address
prefix from one area may be exported to other areas possibly with
summarization or policy-based suppression. A shortest path tree is
constructed at each router for each of the attached OSPF area. IP
datagram destined to other areas is forwarded via the ABR. An OSPF
router that has interface outside of an OSPF domain is called
Autonomous System Boundary Router (ASBR). An ASBR may import
reachability information from outside the AS and IP datagram in and
out of an OSPF domain is via the ASBR.
There have been numerous RFCs and IETF drafts that propose
extensions and enhancement to the RFC2328 where the features include
the following:
1. The OSPF NSSA Option (RFC 1587).
2. OSPF Database Overflow (RFC 1765)
3. The OSPF Opaque LSA Option (RFC 2370)
4. OSPF for IPv6 (RFC 2740)
5. IETF draft, Traffic Engineering Extensions to OSPF
6. IETF draft, OSPF Extensions in Support of Generalized MPLS
In particular, extensions have been added to OSPF to support traffic
engineering for MPLS and GMPLS based networks.
5.1.1 Terminology
ABR Area Border Router. An ABR has interfaces to more than
One OSPF areas.
Area A group of OSPF nodes that inter-connected with each
other and all their interfaces belong to the same OSPF
area.
AS Autonomous System specifies an IP administrative
Bernstein, G. [Page 26]
draft-ipo-optical-inter-domain-00.txt November 2001
domain.
ASBR Autonomous System Boundary Router. An ASBR is an OSPF
node that has interface with routers in other AS, or is
capable of importing IP routes from non-OSPF sources.
Backup Designated Router There is a single BDR elected in OSPF
broadcast or NBMA networks. It takes the
role of the DR when the original DR
fails.
Designated Router There is a single DR elected in OSPF
broadcast of NBMA networks.
LSA Link State Advertisement
NBMA Non broadcast and multi-access network
NSSA An OSPF stub area that allows importing AS external LSA
under a policy-based condition.
Stub Area In OSPF, AS external LSA is flooded throughout all
areas that belong to the same AS. An OSPF area may
optionally be specified as a stub area so that no AS
external LSA flooded into that area.
Virtual Link A logical point-to-point link that connects two
OSPF ABRs.
5.1.2 Neighbor/Adjacency Discovery
The neighbor and adjacency mean different things in OSPF although
both describe the relationship between two inter-connected OSPF
nodes. The inter-connection between OSPF nodes includes both
physical and logical context. In particular, the inter-connection
between two adjacent OSPF neighbors can be realized by either
physical cable or virtual link. The neighbor relationship is at the
IP layer for OSPF. If we can use OSPF or OSPF-like IP based routing
protocol for optical networks, that implies we still need control
plane at the layer-3.
The OSPF neighbor relationship describes the communication using
OSPF Hello protocol between a pair of inter-connected OSPF nodes.
Note the peer relationship is regardless of the underlined interface
type; e.g., on a point-to-point interface, there is a single pair of
OSPF nodes communicating using Hello protocol and on a broadcast
network interface such as Ethernet, the neighbor relationship
applies to each pair of OSPF nodes on that Ethernet. One of the
Bernstein, G. [Page 27]
draft-ipo-optical-inter-domain-00.txt November 2001
usages of the OSPF neighbor relationship is to monitor the bi-
directional communication status between a pair of OSPF nodes,
although on the multi-access networks, the neighbor relationship is
also used for the election of OSPF Designated Router and Backup
Designated Router. Note that this is an "IP layer peer"
relationship. One important item to note is that in the optical
world we typically have much faster and more accurate methods to
monitor the health of the communications between the two nodes. This
is also true for some of the non-optical networks even on the
Ethernet. However, all existing routing protocols (OSPF, PNNI, IS-
IS, etc.) have their own link healthy checking mechanism. What
happens is if the link layer fails, the routing protocol will get
notification with proper handling before its own mechanism kicks in.
But if the failure not due to the lower layer, the protocols
mechanism will be used.
The OSPF adjacency specifies a relationship between a pair of OSPF
nodes that have 2-way communication neighbor relationship as whether
to exchange OSPF link state information between them. Note the
adjacency relationship is always formed between the two inter-
connected nodes on a point-to-point network interface, and is also
so between any pair of nodes on a multi-access network except those
where none of them is either Backup Designated Router or Designated
Router.
The OSPF link state information includes the addresses, nodes,
links, traffic-engineering parameters that belong to all nodes in a
single OSPF area. Information exchanged between separate OSPF areas
is only limited to the addresses. The IETF may decide if and how to
exchange information related to the traffic-engineering parameters
between OSPF areas.
Note the neighbor and adjacency are required elements commonly exist
in any link-state algorithm based routing protocols. Although they
are associated with OSPF nodes at one-hop away (note the hop is
with logical context), the link state information that exchanged
using the protocol mechanism is associated with a much larger
network segment, and that is called Area in OSPF. This is important
since in the optical networks, there might require other neighbor
information and its exchange between inter-connected nodes, such as
those on the same optical ring; it is recommended that separate
protocol(s) be used for that purpose for the reasons with two-fold.
First, it is desirable to separate specific technology such as
SONET, DWDM, ADM etc. from routing protocol that operates at the
networking layer. Second, the information that related to the
underlined technology tends to be with local significance, not
network-wide.
Bernstein, G. [Page 28]
draft-ipo-optical-inter-domain-00.txt November 2001
5.1.3 Addressing & Reachability
OSPF is an interior routing protocol used for IP networks. The OSPF
version 2 supports IPv4 addressing only and OSPF version 3 (work-in-
progress) supports both IPv4 and IPv6 addressing. There is a network
mask associated with an IP address.
In IP networks, an address or address prefix is associated with
entities that include the following:
1. Host-oriented device, including routers, switches,
workstations, etc., and called host address.
2. Network segment called subnet, and called IP subnet address.
3. Network interface that associated with a NE including routers,
switches, etc., and called interface IP address.
OSPF is capable of advertising IP addresses that associated with all
the above. Within a single OSPF area, addresses that associated with
nodes, interfaces, and subnets within that area are advertised
throughout that area with no summarization, along with other link
state information. This is important such that routing path for any
reachable address can be calculated in very optimized fashion.
On the other hand, OSPF is capable of summarizing addresses that
associated with one area before advertising them to other areas. In
addition, OSPF can summarize addresses and selectively (based on
administrative policies) advertising them from one Autonomous System
(AS) to others via inter-domain routing protocols such as BGP-4.
For reachable addresses across boundaries topology of OSPF areas or
AS, the routing path for them may or may not be optimized due to the
lack of information since the OSPF link state advertisements not
across those boundaries.
5.1.4 Topology Discovery & Dissemination
The information that contained in the link state advertisements
includes three categories as follows:
1. Reachability of IP addresses and IP subnets (discussed in the
Section 2.1.2)
2. Traffic engineering parameters (discussed in the Section 2.1.4)
3. Network topology including OSPF nodes, links and their
connectivity (discussed in this section)
The topology related link state advertisements only flow within a
single OSPF area but not beyond. In a single OSPF area, each node
knows all the other nodes, their links and connectivity without any
aggregation. This is the basis for routing path optimization within
a single OSPF area.
Between OSPF areas, there is no topology related link state
advertisements exchange. The only information across the OSPF area
boundary is the reachability of the IP addresses and IP subnets via
OSPF Area Border Routers. This scenario is also true across OSPF
domains where the reachability is via OSPF Autonomous System
Bernstein, G. [Page 29]
draft-ipo-optical-inter-domain-00.txt November 2001
Boundary Routers. Therefore, while OSPF is a link-state algorithm
based routing protocol, it is only true within an OSPF area and at
the next higher level of the hierarchy it can only perform distance
vector algorithm based routing. This will most likely be one of the
major shortcomings for OSPF to be used for optical networks where if
an optical network needs to scale to very large and be partitioned
into OSPF areas, there will be no topology information at all across
the area boundary.
5.1.5 Resources
The OSPF was originally defined for routing IP packets with best
effort. The MPLS/GMPLS extensions to the OSPF add the ability for
OSPF to advertise traffic engineering parameters using opaque Link
State Advertisements.
The traffic engineering parameters advertised by OSPF are associated
with OSPF links that can carry user traffic. Currently the set of
engineering parameters including the following:
1. Maximum link bandwidth
2. Available link bandwidth (with 8 priority levels)
3. Traffic engineering metric
4. Administrative class
5. Protection type
6. Shared risk link group
Like the scope of the topology link state advertisements, the
advertising for traffic engineering information is also restricted
within a single OSPF area. This makes the constraints based routing
using OSPF very effective within a single OSPF area but not beyond.
This is another major shortcoming of OSPF that needs to be
considered for using OSPF in the optical networks.
5.1.6 General Protocol Properties
This section briefly discusses the general protocol properties of
the OSPF in the specific areas as in the following sections.
5.1.6.1 System Overhead
The memory required to support OSPF operation is a variable
depending on several factors including the following:
1. Whether the node is an ABR or not ABR usually consumes more
memory since it needs to maintain database for multiple OSPF
areas.
2. Whether the node is an ASBR or not - ASBR may also consume more
memory since it needs to store routes obtained from other
routing protocols such as BGP-4.
3. The number of nodes and links within each OSPF area.
4. The number of reachable addresses within a single OSPF area,
imported from other areas and other AS.
5. The amount of the traffic engineering information within an
OSPF area.
Bernstein, G. [Page 30]
draft-ipo-optical-inter-domain-00.txt November 2001
Most of the OSPF routing messages are re-transmittable driven by the
associated timers in the order of seconds and the timer values are
configurable.
5.1.6.2 Network Resource Overhead
OSPF routing messages can either be carried in-band, out-band or
out-of-fiber. The consumption of the network resources depend on
factors including the following:
1. The amount of the routing messages that need to be put on the wire
depends on factors that similar to those listed above (Section
2.1.7.1) for the memory consumption.
2. Interface type it is visible that during transit time (election
of (Backup) Designate Router, topology change, etc.), the OSPF
routing messages on interfaces of multi-access networks (Ethernet,
etc.) are very large and the actual amount depending on the
implementation, the number of routers on the subnet, etc.
3. OSPF packets
Most of the OSPF packets as defined in the RFC 2328([OSPF1])are
very compact that saves the bandwidth consumption on the wire.
The OSPF address advertising LSA (type 3 and 5) only allows to
contain one single reachable address. This is not economic at
all.
Some of the traffic engineering parameters contained in the
OSPF opaque LSA are not with compact format such as bandwidth
based on priority.
Consideration needs to be taken in this regard since in the optical
networks, control channels are usually out-of-band or out-of-fiber,
where there may not always be large quantity of bandwidth available
for routing messages.
5.1.6.3 Reliability
OSPF does not have re-start capability originally other than re-
build its routing database from scratch upon re-start. This fact
cannot be accepted by most of the carrier-class networks today.
Note however, that routing protocols for the optical network are not
service impacting in the way that they are for IP datagram routing.
Hence, for connection/circuit based networks, the routing protocols
re-start capability may not be critical as in the datagram networks.
The only scenario is if we use signaling protocol to setup
Bernstein, G. [Page 31]
draft-ipo-optical-inter-domain-00.txt November 2001
circuits, and if we also (sometimes) have to use the signaling
protocol to re-route failed circuits, there will be a busy routing
scenario where the routing protocols re-start appears important.
However, the IETF has a few proposals to improve this and in
particular the proposal as described in the [OSPF5] allows a
graceful re-start along with backward compatibility.
5.1.7 Scaling Capability
OSPF is capable of scaling and can handle large-scale IP networks.
The scalability of a routing protocol is achieved by reducing the
amount of the routing information exchanged between network
segments. The reduction of the routing messages will certainly have
impact on the routing efficiency and optimization. The kind of
information reduction and their impact for OSPF are as follows:
1. Address summarization as described in the Section 2.1.2, OSPF
is capable of performing address summarization at the area
boundary and Autonomous System boundary. As a result, the
amount of the addresses that advertised throughout the IP
networks is greatly reduced. The reduction of the address
advertisement does not affect the reachability since the
routing path across OSPF areas and Autonomous Systems is via
OSPF Area Border Router and Autonomous System Boundary Router,
respectively.
2. Topology aggregation as described in the Section 2.1.3, OSPF
is capable of topology aggregation, but to the extreme, i.e.,
there is no topology information exchange between OSPF areas
and Autonomous Systems at all. This greatly reduces the amount
of the advertisements throughout the IP networks but with a big
price, i.e., routing path across the OSPF areas or Autonomous
Systems will most unlikely be optimized.
3. Traffic engineering information aggregation as described in
the Section 2.1.4, there is no traffic engineering information
exchange between OSPF areas and Autonomous Systems as defined
in current OSPF TE extensions. As a result, the scaling purpose
is achieved but the routing with traffic engineering
requirements across OSPF areas or Autonomous Systems is very
inefficient or ineffective.
4. OSPF has another feature that limits the flooding of the Link
State Advertisements (LSA) an OSPF area can be configured as
a so-called Stub Area and as such no AS-based LSA is allowed to
flow into it. The AS-based LSA includes the OSPF type 5 LSA
only as today. The OSPF type 5 LSA contains reachable addresses
from other Autonomous Systems; the communication between other
AS and a stub area is via the ABR(s) that attach to the stub
area.
The address summarization capability of OSPF is very valuable. The
topology aggregation behavior of OSPF does not sound great in multi-
Bernstein, G. [Page 32]
draft-ipo-optical-inter-domain-00.txt November 2001
area operation especially for traffic engineering based
applications.
5.1.8 Interworking Capability
OSPF cannot interwork with any other protocol using any common
protocol messages. However, OSPF as an IP routing protocol is
capable of interworking in terms of sharing and exchanging IP
reachability information as follows:
1. OSPF can operate together with any other IP routing protocols
on the same router/switch where the IP forwarding table (FIB)
can be shared among all the IP routing protocols.
2. OSPF is capable of exchanging IP reachability information with
any other IP routing protocols. The most common scenario is the
reachability information exchange between OSPF and BGP-4.
The NNI routing protocol, if not OSPF, needs to have the capability
of interworking with OSPF and IS-IS where the minimum requirement is
the exchange of reachability information, although the exchange of
topology and resource information are also highly desired.
5.1.9 References
[OSPF1] RFC 2328, OSPF Version 2.
[OSPF2] IETF draft, Traffic Engineering Extensions to OSPF, draft-
katz-yeung-ospf-traffic-05.txt
[OSPF3] IETF draft, OSPF Extensions in Support of Generalized
MPLS,
draft-ietf-ccamp-ospf-gmpls-extensions-00.txt
[OSPF4] RFC 2370, The OSPF Opaque LSA Option
[OSPF5] IETF draft, Hitless OSPF Restart, draft-ietf-ospf-hitless-
restart-01.txt
[OSPF6] IETF draft, OSPF Version 2 Management Information Base,
draft-ietf-ospf-mib-update-06.txt
[OSPF7] RFC 1587, The OSPF NSSA Option
5.2 IS-IS and Integrated IS-IS
IS-IS was originally developed at Digital as Decnet Phase V
IS-IS is a link state routing protocol originally specified to route
OSI CNLS packets between Intermediate Systems (ISOs description for
routers). The specification for the protocol can be found in ISO
document 10589.
RFC 1195 defines Integrated IS-IS, an adaptation of this protocol
capable of routing both IP and OSI packets.
5.2.1 Terminology
IS-IS Intermediate System to Intermediate System
routing exchange protocol
Intermediate system ISO term for a router
Bernstein, G. [Page 33]
draft-ipo-optical-inter-domain-00.txt November 2001
CLNP Connection-Less Network Protocol ISOs
version of IP
LSP Link State PDU (Protocol Data Unit)
ISO International Standards Organization
SPF Dijkstras Shortest Path First algorithm
5.2.2 Neighbor/Adjacency Discovery
Neighboring routers are discovered through the periodic transmission
and reception of IS-IS Hello packets. The hello packet provides the
neighbor with the routers network layer address and a holding time.
The holding time is the number of seconds that the neighbor should
maintain reachability to the sending router without receiving
further IS-IS Hello packets. The Hold timer is set so that even if
some Hello packets are dropped, the neighbor connection remains
active. One version of the IS-IS Hello packet is used for point to
point links, the other is used for the LAN version, this contains
additional information such as the ID of other routers to ensure
that connectivity between neighbors is bi-directional.
5.2.3 Addressing & Reachability
Integrated IS-IS supports NSAP, IPv4 and IPv6addressing. In IS-IS
routing, the network is partitioned into routing domains
(equivalent to AS in OSPF). Routing domain boundaries are defined by
network management setting some links to be exterior links. If a
link is marked as exterior, no IS-IS routing messages are sent on
that link.
5.2.3.1 CLNP/NSAP addressing and routing
SDH/SONET, DECnet Phase V, ATM and cellular digital packet data use
CLNP Addressing, sometimes known as NSAP addressing. Routing
boundaries between routers in an area, areas in a domain and between
domains are determined by sections of the NSAP address. A router
is any entity that routes CLNP packets on the basis of their layer 3
addresses, therefore a router in this context can be a dedicated
CLNP router or a controller card in a SONET ADM.
ISO 10589 specifies that up to three areas addresses may be
configured within an area. A number of implementations allow more
Bernstein, G. [Page 34]
draft-ipo-optical-inter-domain-00.txt November 2001
than that. In order for two neighboring level one routers to be in
the same area any one of the area addresses must match any one of
the areas in a received packet.
The NSAP address contains an area section and an ID section; for
routers to be within the same area, the area portion of their NSAP
address must to be identical. Level 1 routes on the basis of the ID
field, there is no hierarchy in the ID portion of the address. The
ID field is assumed to be a flat address space with no topological
significance. Level 2 routes according to the longest prefix of the
area portion of the address. Level 2 routing is similar to IP
routing.
In order to support end OSI end systems, a router must run the ES-IS
protocol. This is not the same as IS-IS.
OSI IS-IS routing makes use of a two-level hierarchical routing.
Level 1 routers know the topology in their area, including all the
routers and end systems in their area. However the level 1 routers
do not know the identity of routers outside their area. Level 1
routers forward all traffic outside their area to a level 2 router
in their area. Level 2 routers do not need to know the topology
within any level 1 area, except that in some cases a level 2 router
may also be a level 1 router.
5.2.3.2 IP addressing and routing
Level 1 routers within an area exchange LSPs that identify the IP
addresses that are reachable by each router. Specifically zero or
more [IP address, subnet mask, metric] combinations may be included
in each LSP. A level 1 router routes as follows:
1. If a specified destination address matches an [IP address,
subnet mask, metric] reachable within the area, the packet is
routed using level 1 routing.
2. If a specified destination address does not match any [IP
address, subnet mask, metric] combination listed as reachable
within the area, the packet is routed towards the nearest
level2 router. Provided that the level two router is attached
to another area, otherwise this does not hold.
Level 2 routers include in their level 2 LSPs a complete list of [IP
address, subnet mask, metric] specifying all IP addresses reachable
in their area. This information may be obtained from a combination
of the level 1 LSPs (obtained from level 1 routers in the same
area), and/or by manual configuration. In addition, level 2 routers
may report external reachability information, corresponding to
addresses thatcan be reached via routers in other routing domains.
Bernstein, G. [Page 35]
draft-ipo-optical-inter-domain-00.txt November 2001
Some implementations allow this with level one LSPs as well as level
two. If supported the option is configurable.
5.2.4 Topology Discovery & Dissemination
Topology discovery and dissemination is achieved by means of LSPs,
Link State PDUs. Each router constructs a packet known as a link
state packet or LSP (OSPF uses the term Link State Advertisements or
LSAs) which contains a list of reachable address prefixes, areas and
routers. An IS-IS router will generate a LSP periodically or when
it has a new neighbor, the cost of the link to an existing neighbor
has changed or if the link to a neighbor has gone down. The LSP is
then transmitted to all other routers. Each router stores
information gained from the most recently generated LSP from the
other routers. The LSPs provide each router with a complete map of
topology. LSPs are only exchanged between routers of their own type
(i.e. level 1 or level2), though as stated earlier, some routers can
operate as level 1 and 2 routers.
Based on the topological information obtained from LSPs, each router
can use SPF to calculate the optimal (least cost) path to a
destination. However whilst an optimal route can be chosen within
an area, the extent to which an optimized path is chosen between
areas is dependent on the level of summarization used for address
prefixes in level 2 routing tables. As with OSPF, there is a
balance between reduced table entries (to minimize router memory
usage)and an optimum path route.
5.2.5 Resources
Work is currently being undertaken at the IETF to add extensions to
Integrated IS-IS LSPs to provide the ability for IS-IS to advertise
traffic engineering parameters, in parallel to similar efforts for
OSPF.
A set of parameters taken from the draft-ietf-isis-gmpls-extensions-
04.txt Internet Draft, includes the following:
1. Maximum link bandwidth
2. Reservable link bandwidth
3. Unreserved bandwidth
4. TE Default metric
5. Link Protection Type
6. Interface Switching CapabilityDescriptor
There are many other extensions being proposed in particular those
in support of circuit provisioning of SDH, SONET and G.709
equipment.
Bernstein, G. [Page 36]
draft-ipo-optical-inter-domain-00.txt November 2001
5.2.6 Interface types and network Medium Support
Whilst Integrated IS-IS can route both CNLP and IP packets, the
routing protocol itself is based on CNLP. This requires that all
infrastructure used in an IS-IS routing domain be able pass IS-
IS/CNLP packets. In practice CNLP can be carried on a variety of
data-link layer technologies such as Ethernet (all types), Token
Ring, FDDI, and Frame Relay. Also POS in which case OSICP is used.
5.2.7 General Protocol Properties
This section briefly discusses the general protocol properties of
Integrated IS-IS.
5.2.7.1 System Overhead
The amount of system overhead (memory and processing)required for
Integrated IS-IS or IS-IS routing is dependent on the following:
-Whether the router participates in both level 1 and level 2
routing a level 1/ level 2router will need to maintain a detailed
database for its own area and a summarized database for all other
areas.
-The number of nodes within each IS-IS area.
-The total number of reachable addresses in every IS-IS area-thus
affecting the size of level 2 routing tables.
-The level of address summarization used in level 2 routing. As
mentioned before, this summarization is done at the expense of
calculating optimal path routes.
-The amount of the traffic engineering information disseminated
within an IS-IS area. GMPLS provides for the dissemination of
resource information using IS-IS extensions. Such information will
typically include available ports, timeslots (SONET) and wavelengths
(Optical layer equipment). The level churn of connections will have
a serious impact on the amount of system overhead required to
process such IS-IS information particularly if such information is
to be processed in real-time. Such concerns are equally applicable
to OSPF extensions.
5.2.7.2 Network Resource Overhead
IS-IS routing messages can either be carried in-band, out-band or
out-of-fiber. From the perspective of Transport network elements
under consideration in the OIF, in-band could mean the DCC bytes
available within SONET and SDH overheads. This is currently how
standards-based SDH and SONET configuration messages and alarms are
Bernstein, G. [Page 37]
draft-ipo-optical-inter-domain-00.txt November 2001
routed between network elements and their respective element
managers. Out-band routing messages could be carried on a
supervisory wavelength, whilst out-of-fiber routing messages could
be carried out on a physically independent network such as an
Ethernet network.
Where out-of-fiber networks are used to transport routing messages,
such networks often rely on routers to route packets end-to-end.
Unless a different routing instance or static routing is used, the
topology information that is shared will include both the topology
of the transport network and the out-of band network. This has an
impact on the amount of routing information that passes through
transport network elements where both in-band and out-of-band
resources are used.
In-band channels such as SDH/SONET DCC channels represent free
resource, however such resources are scarce(192kbps and 576kbps for
Section and Line DCC) and it is important to ensure that unnecessary
consumption is avoided at all costs.
Consumption of the network resources depend on factors including the
following:
-The amount of the routing messages that need to be put on the wire
depends on factors that similar to those listed above (Section
2.2.7.1).
-IS-IS LSPs containing traffic engineering parameters are likely
consume a large amount of bandwidth. Work needs to be done to
quantify this. However whilst IS-IS is an improvement over OSPF, in
that a single LSP can contain multiple reachable addresses, it is
the total bandwidth that matters. The total amount of information
carried in all the LSPs, LSP count in itself is not informative. Im
not sure that I would say that IS-IS represents an improvement over
OSPF here. IS-IS is limited in that a single logical LSP can at most
be composed of 256 physical LSPs, each constrained to the local MTU.
This limitation makes it impossible for IS-IS to carry the number of
routes that OSPF can support. Furthermore, on a LAN, the DR in IS-IS
will be sending out a Complete Sequence Number Packet (CSNP) every
second or so. Actually this CSNP may well require multiple CSNPs if
there are a large number of LSPs in its database. So OSPF is more
efficient in terms of its use of bandwidth on the wire. However that
efficiency comes at a price. OSPF is a more complex protocol.
Bernstein, G. [Page 38]
draft-ipo-optical-inter-domain-00.txt November 2001
5.2.7.3 Reliability
IS-IS in a similar manner to OSPF does not have re-start capability,
other than re-build its routing database from scratch upon re-start.
This fact cannot be accepted by most of the carrier-class networks
today. When after a significant failure within a network, a large
number of network elements are brought back on-line the exchange
of LSPs could exceed the traffic limits of the network carrying the
LSPs. Such congestion can be caused by insufficient link capacity
or intermediate router throughput capability. The result of such
congestion can be the prolonged time required to populate route
tables (through loss of LSPs), or even a complete failure to re-
establish connectivity.
Such failures are inherent in all networks that use Link State
Routing Protocols, and work is being done in the IETF and ATM Forum
to modify IS-IS, OSPF and PNNI to ensure that the appropriate
mechanisms are in place to prevent such occurrences. This will
ultimately have an impact on scalability.
IS-IS has already in place a mechanism to deal with router
congestion (OSPF has something similar as an option). When the IS-IS
routing database exceeds its allocated memory, a router that cannot
fit a new LSP into its database refuses to acknowledge the LSP. The
neighbor continues trying to transmit the LSP. A router that is
forced to refuse an LSP sets a flag in its LSP indicating that its
LSP database is full. Other routers use paths through that router
only if no path through non-overloaded routers exits. This behavior
means that a temporary overload situation heals itself without human
intervention.
5.2.8 Scaling Capability
Scaling issues have already been addressed in other parts of this
document. The scaling capability of IS-IS is no different to OSPF,
except in one regard: See my comment above. OSPF can carry more
routes than ISIS.
The large address space and hierarchical structure of the NSAP
format often used by IS-IS routers (20 bytes versus 4 bytes forIPv4)
allows the creation of worldwide networks without the need for
address translation or private addressing.
5.2.9 Interworking Capability
Integrated IS-IS can route between network elements regardless of
whether IP or NSAP addressing is used. Where it becomes necessary
to connect IP network elements to NSAP based network elements,
Integrated IS-IS provides an easy way of doing so. Information can
be learnt from other Interior Gateway Protocols such as RIP and OSPF
(sometimes known as route redistribution).Interworking with inter
domain routing protocols is described in the next section.
Bernstein, G. [Page 39]
draft-ipo-optical-inter-domain-00.txt November 2001
5.2.10 Inter Domain Routing Protocols
Inter domain routing protocol information (IRPI) is included in
level 2 LSPs and serves to provide information only to inter domain
routing protocols. The IRPI field allows interdomain routers to find
each other. The interdomain routers within a routing domain could
discover which of the level 2 routers within the domain could
discover which of the level 2 routers within the domain were also
interdomain routers.
Integrated IS-IS allows IP addresses reachable via inter-domain
routing to be reported in level 2 LSPs in the IP external
reachability information field. This includes routes learned from
OSPF, RIP or any other external routing protocol.
5.2.11 References
-RFC 1195, Use of OSI IS-IS for Routing in TCP/IP and Dual
Environments
-Radia Perlman, Interconnections
-John T. Moy,OSPF, Anatomy of an Internet Routing Protocol
-IETF Internet Draft, draft-ietf-isis-gmpls-extensions-04.txt
-ISO 10589 Intermediate system to intermediate system Intra-Domain
routing information exchange protocol for use in conjunction with
the protocol for providing the Connectionless-mode Network Service
5.3 BGP
From RFC1771:
"The primary function of a BGP speaking system is to exchange
network reachability information with other BGP systems. This
network reachability information includes information on the list of
Autonomous Systems (ASs) that reachability information traverses.
This information is sufficient to construct a graph of AS
connectivity from which routing loops may be pruned and some policy
decisions at the AS level may be enforced."
5.3.1 Terminology
Autonomous System
BGP Speaker
External Peer
Internal Peer
NLRI Network layer reachability information
5.3.2 Neighbor/Adjacency Discovery
BGP does not include a neighbor discovery protocol. In fact BGP
uses the term peer since communication is not always limited to
directly attached neighbors. All peers must be configured. As an
inter-domain protocol this makes sense, since you do not necessarily
want any two boxes that get connected together in differing domains
Bernstein, G. [Page 40]
draft-ipo-optical-inter-domain-00.txt November 2001
to start exchanging routing information. Two types of peers are
distinguished in BGP. An external peer is a BGP speaker in a
different AS, while an internal peer is a BGP speaker within the
same AS. There is however a working assumption that external peers
are "directly connected" however this is not strictly required.
Internal peers are frequently not directly connected.
5.3.3 Addressing & Reachability
BGP works with IPv4 and IPv6 addresses. Much of BGP's strength lies
in its very general mechanisms for dealing with reachability
information. In particular this information can be aggregated
(summarized), filtered on input, and filtered on output (for
controlled information dissemination).
BGP is "the" reachability protocol. The Update message contains a
path (AS_PATH) that furnished at least one possible route to reach
the destinations summarized (via prefixes) in the Network Layer
Reachability Information (NRLI) field. Note that the NEXT_HOP
attribute can be used in terms of the next optical hop (rather than
IP hop). Hence as it stands BGP can be used for arbitrary optical
reachability. The BGP sessions are set up via the IPCC addresses (IP
routable) but the information exchanged pertains to the optical
network not the IP control channel network.
BGP-4 [BGP1] provides a number of policy mechanisms that relate to
how routing information is used and disseminated. In particular the
E-BGP border router model keeps distinct the routing information
received from each of a border routers autonomous systems external
peers (Adj-RIBs-In -- Adjacent Routing Information Base In), the
routing information that the Autonomous System (AS) itself is using
(Loc-RIB -- Local Routing Information Base), and the routing
information that the AS forwards onto its external peers (Adj-RIBs-
Out -- Adjacent Routing Information Base Out). Via this model one
can develop policies with regards to which routes get chosen for use
in the AS, i.e., which routes from the Adj-RIBs-In are chosen to
populate the Loc-RIB. One also develops policies concerning what
routing information gets advertised to external peers, i.e., which
routes from Loc-RIB gets exported to each of the Adj-RIBs-Out.
The choice of which routes get imported for local routes generally
is concerned with the "quality" of those advertising the routes
since not too much else is known (besides the AS path vector). In
deciding which routes to advertise to external peers "transit
policies", i.e., whose traffic is allowed to transit this AS is the
prime consideration.
In the MPLS and in particular the explicitly routed optical case we
have a very strong additional policy mechanism, that of connection
admission control (CAC). Although an optical AS probably shouldnt
advertise transit capabilities that it doesnt wish to support, CAC
during connection establishment will be the final arbiter of any
Bernstein, G. [Page 41]
draft-ipo-optical-inter-domain-00.txt November 2001
transit policy. In addition, some areas that are being addressed by
policies in the IP datagram case such as load balancing are much
easier to implement via CAC and/or explicit routing.
5.3.4 Topology Discovery & Dissemination
BGP is a path vector type of protocol which can be thought of as a
variant of a vector distance protocol. For each set of NLRI a path
vector is kept to indicate the sequence (or collection) of
autonomous systems that are traverse to reached these summarized
destinations. The main reason for keeping this information is to
avoid routing loops. Although the path vector information can give
us some notion of AS connectivity this is by necessity extremely
limited. As part of the BGP route selection process only one route
to a set of destinations can be selected and only routes that are
actually used can be advertised via BGP to other ASs.
5.3.5 Scaling Capabilities
BGP-4 was originally defined as an inter-domain routing protocol for
traditional IP networks where a common scenario was the BGP-4 would
be used as point-to-point communication protocol between ASBR that
is not very large in number in the IP networks. When the BGP-4 is
used in other scenario where the situation does not hold, as for
example when BGP-4 is used in a BGP/MPLS based VPN network, there
appears to be a serious scaling problem, i.e., there tends to be a
large number of end-to-end BGP-4 sessions in a network.
There exists mechanism to resolve the scalability problem using BGP
reflectors ([BGP4]). However, BGP reflector uses a server-client
model that creates a single-point-failure problem.
5.3.6 Interworking Capability
BGP there exist several implementations where BGP interoperates with
popular IGPs (Interior Gateway Protocols) such as OSPF and IS-IS.
5.3.7 References
[BGP1] Rekhter Y., and T. Li, "A Border Gateway Protocol 4 (BGP-4)",
RFC 1771, T.J. Watson Research Center, IBM Corp., cisco Systems,
March 1995.
[BGP2] Rekhter, Y., and P. Gross, "Application of the Border Gateway
Protocol in the Internet", RFC 1772, T.J. Watson Research Center,
IBM Corp., MCI, March 1995.
[BGP3] Rekhter, Y., Rosen E., Carrying Label Information in BGP-4,
RFC 3107
[BGP4] Bates, T. et al, BGP Route Reflection An Alternative to
Full Mesh IBGP, RFC 2796
5.4 PNNI routing
PNNI standards for Private Network-to-Network, or Node-to-Node
Interface used for ATM networks. PNNI routing protocol is defined by
Bernstein, G. [Page 42]
draft-ipo-optical-inter-domain-00.txt November 2001
the ATM Forum and is used for ATM networks. PNNI routing protocol is
link state algorithm based and in fact many elements are similar to
other routing protocols such as OSPF for IP networks and IS-IS for
OSI networks. PNNI routing protocol is capable of scaling to very
large networks through its hierarchical architecture. The addressing
used by PNNI routing protocol is called ATM End System Address, or
AESA, that is similar to the NSAP addressing.
The companion signaling protocol used in PNNI network is the PNNI
signaling protocol that is similar to the UNI4.0 signaling protocol
but with symmetric interface, among other differences. PNNI
signaling protocol is also defined by the ATM Forum and actually
contained in the same specification as the PNNI routing protocol.
5.4.1 Terminology
ABR Available Bit Rate
Adjacency The relationship between two PNNI communicating
neighboring peer nodes.
AESA ATM End Station Address
Aggregation Token A number assigned to an outside link by the
border nodes at the ends of the outside link.
The same number is associated with all uplinks
and induced uplinks associated with the outside
link. In the parent and all higher-level peer
group, all uplinks with the same aggregation
token are aggregated.
Border Node A logical node that is in a specified peer group, and
has at least one link that crosses the peer group
boundary.
CBR Constant Bit Rate
Complex Node Representation A collection of nodal state parameters
that provides the detailed state
information associated with a logical
node.
DTL Designated Transit List
Entry Border Node The node that receives a call from anoutside
link. This is the first node within a peer group
to see such a call.
Exit Border Node The node that will progress a call to an outside
link. This is the last node within a peer group
to see such a call.
Bernstein, G. [Page 43]
draft-ipo-optical-inter-domain-00.txt November 2001
Horizontal Link A link that is between two PNNI nodes within the
same peer group.
Inside Link Synonymous with horizontal link.
Level Indicator PNNI hierarchical level indicator, from 0 to 104
inclusively.
LGN Logical Group Node
Outside Link A link to a lowest level outside node. In
contrast to an inside link or an uplink, a
outside link is not included as part of the
topology information.
PTSE PNNI Topology State Element
Peer Group A set of logical nodes which are grouped for
purposes of creating a routing hierarchy.
PGL Peer Group Leader
RCC Routing Control Channel
Service Category Specify the CBR, real-time VBR, non real-time
VBR, ABR, or UBR.
Simple Node Representation An aggregation at the maximum level for
a logical node by only advertising the
logical node itself, in the contrast of
the complex node representation.
SVCC ATM SVC-based Connection
UBR Unspecified Bit Rate
Uplink Represents a connectivity between a border node and an
up node.
VBR Variable Bit Rate
VPC ATM Virtual Path Connection
Up node The node that represents the outside neighbor of a
border node in the common peer group. The upnode must
be a neighboring peer of one of the border nodes
ancestor node.
5.4.2 Neighbor/Adjacency Discovery
The neighbor relationship specifies when two PNNI nodes that can
communicate with each other directly on a PNNI link. The adjacency
Bernstein, G. [Page 44]
draft-ipo-optical-inter-domain-00.txt November 2001
specifies the relationship between two PNNI neighboring peer nodes
or peer groups. A PNNI link can be one of the three as follows:
1. A physical link that inter-connects two ATM switches.
2. A VPC-based PNNI logical link.
3. A SVCC-based PNNI logical link.
Note in PNNI networks, all inter-connections between PNNI nodes are
point-to-point. All PNNI routing messages are exchanged between PNNI
neighbors. PNNI routing messages are carried on the Routing Control
Channel (RCC) that is out-of-band but always in the same fiber where
the data channels reside. Note that the definition of "out of band"
in the packet switching case is a bit different then in circuit
switching case.
A Hello protocol is used to monitor the bi-directional communication
status between a pair of PNNI nodes, determines if the two neighbors
belong to the same peer group or not, etc. If the two neighbors
belong to the same peer group, the link is called Inside Link
otherwise Outside Link.
PNNI neighbors on an Inside Link will also exchange PNNI topology
information in packets called PNNI Topology State Element (PTSE).
PTSE carries the topology information including nodes, links,
addresses, and resources, etc. within that peer group. The operation
model of a PNNI peer group is similar to that of an OSPF area
wherein all routing information within that peer group is known to
all the nodes in that peer group. In addition, a peer group elects
one single node as the Peer Group Leader node. A separate PNNI node
instance at the higher hierarchical level may run on the same ATM
switch platform as the PGL of the lower hierarchical level resides.
The PNNI node instance at higher levels is called Logical Group Node
(LGN). The LGN then forms a neighbor relationship with any other LGN
in other peer groups on a SVCC-based RCC that allows routing
messages including Hello exchanged between the two LGNs on that RCC.
This scenario is recursive in nature.
PNNI neighbors on an Outside Link will also exchange hierarchical
information if available, including the identity of the higher level
PNNI nodes, peer groups, etc. PNNI nodes with Outside Links are
called Border Nodes.
One obvious similarity between ATM networks and optical networks is
the circuit-based interconnection. The other similarity is that both
networks dedicated to a single underlined networking technology,
i.e., ATM and optics, respectively. The similarity may be useful for
us to realize that as one part of the networking mechanism - the
routing protocol is also possible to share much similarity.
Note however "optics" is actually quite varied but it we want to
separate out technology specific routing information from more
general information. What is similar across all the optical
technologies that we will be dealing with is that they are circuit
Bernstein, G. [Page 45]
draft-ipo-optical-inter-domain-00.txt November 2001
oriented. This tends to make bandwidth accounting/allocation quite
simple hence we spend almost no time worrying about it (except for
compact representations on the wire) while ATM spent a lot of time
on packet QoS issues.Technology-specific routing information (such
as regeneration requirements in all-optical networks may create
essentially different QoS issues )
5.4.3 Addressing & Reachability
PNNI is an interior routing protocol used for ATM networks. The
addresses used in PNNI networks are ATM End System Address (AESA).
The AESA is 20-byte in length. The common portion of several AESAs,
consisted of the most significant bit stream of these AESA is called
ATM address prefix. The ATM address prefix is used to summarize
AESAs and other address prefixes.
An AESA is associated with an ATM host. In PNNI network, each PNNI
node has a unique AESA in the associated routing domain. Other ATM
devices register their AESAs with their attached ATM switches via
the ILMI protocol ([2]), and these AESAs are then advertised by the
associated PNNI node at the lowest hierarchy on that ATM switch
throughout the associated PNNI peer group along with the identity of
the advertising node itself. The advertised AESAs are therefore
reachable by other nodes in the same peer group.
The AESAs that registered with a PNNI node may be summarized if
possible and in this case, only the ATM address prefix is
advertised. This is one addressing scaling mechanism in the PNNI
networks.
All the AESAs and ATM address prefixes belong to a single PNNI peer
group are stored on each node in that peer group. The PGL in that
peer group may summarize these AESAs and address prefixes, then
deliver them to the LGN at the next higher hierarchical level and
advertise them to the LGNs peer nodes and peer groups, such that
these AESAs and address prefixes are reachable from other peer
groups. Note the PGL and LGN perform a special role here as passing
address reachability information along the PNNI hierarchical levels,
but the actual data communication between peer groups is via the
border nodes.
Currently there does not exist any ATM inter-domain routing protocol
and as such, reachability for AESAs and address prefixes can only
possible via static configuration.
ATM address follows the same format of NSAP in the sense of 20-byte
in length, although there are different encoding rules for ATM
addresses (NSAP format, DCD format, etc.). PNNI routing protocol
itself does not have the ability to carry IP address but there
exists a PNNI Augmented Routing (PAR), also defined by the ATM
Forum, is capable to do so. But I dont think any vendor implemented
it yet. PAR was expected to be implemented on ATM switch and edge IP
Bernstein, G. [Page 46]
draft-ipo-optical-inter-domain-00.txt November 2001
router so in the ATM core, it runs PNNI for ATM networks but
interface with IP routers for IP routing.
5.4.4 Topology Discovery & Dissemination
The routing information is contained in the PNNI Topology State
Element (PTSE) that includes three categories as follows:
1. Reachability of AESA and ATM address prefixes (discussed in
the Section 2.4.3)
2. Traffic parameters as called resource available information
(discussed in the Section 2.4.5)
3. Network topology including PNNI nodes, links, their
connectivity and hierarchy (discussed in this section)
The nodes and links within a single PNNI peer group are advertised
throughout that peer group. Note this behavior is the same for all
the peer groups - at each of the hierarchical levels. As similar to
the OSPF area, routing path within a single PNNI peer group can be
made very optimized Within a single peer group, the detail topology
information including nodes, links and addresses along with
resources information are known by all nodes, so the constraint-
based routing can take such advantage to make the routing path very
optimized. Note the routing messages are only exchanged on Inside
Links.
There are two types of PNNI nodes, i.e., the LGN/PGL nodes and the
border nodes, the LGN/PGL need to perform additional duties as
described as follows.
Each LGN at higher level hierarchy needs to pass the topology
information down to the PNNI node at the next lower hierarchy, i.e.,
the PGL of the peer group at that lower level. The PGL then
advertises that information throughout its peer group. This behavior
is recursive in nature.
Each PNNI border node while performing Hello with its neighbors
belong to other peer groups and in addition, it exchanges the
identity of its ancestor nodes, i.e., LGNs and peer groups at higher
hierarchies with its peers. The obtained information is advertised
via a so-called Uplink PTSE throughout the border nodes own peer
group. The PGL in that peer group will then pass the information
contained in the Uplink to one or more of the LGNs at its higher
level hierarchies. The information contained in the Uplink will be
used to identify the neighboring LGN, if identified, a PNNI logical
link will be established between the two LGNs at that higher
hierarchical level. Note while the logical link is one-hop away, it
is based on SVCC-based RCC such that physically it may contain
multiple physical hops. A logical link logically inter-connects two
adjacent peer groups, and it is advertised within the associated
peer group at the higher level hierarchy as well as passes down to
the peer groups below that hierarchy, so that PNNI nodes at lower
Bernstein, G. [Page 47]
draft-ipo-optical-inter-domain-00.txt November 2001
level hierarchies understand the connectivity with other peer groups
also. This behavior is also recursive in nature.
As described above collectively, the topology information of a
single peer group is always exactly known by all nodes in that peer
group, but the topology information of other peer groups is not
known except the inter-connectivity between peer groups.
Therefore, routing path within a single peer group is possibly made
very optimized, but not across peer group boundaries. I'm not sure
that this is that bad, at least one gets to see how the peer groups
are connected and choose from there. This is much better than not
having that visibility. But of course not as optimal as complete
information.
5.4.5 Resources
ATM is well known for its QoS and traffic management capability, and
the PNNI routing protocol is also capable of advertising various of
traffic engineering parameters, as so-called Resource Availability
information.
The Resource Available information is contained in the following
PTSEs:
1. Reachable address PTSE
2. Horizontal links PTSE advertisements for PNNI Inside Links
and logical links
3. Uplinks PTSE
The set of traffic parameters currently advertised by PNNI routing
protocol is service category based and directional based, and as
follows:
1. Maximum cell rate (MaxCR)
2. Available cell rate (AvCR)
3. Administrative weight (AW)
4. Cell transfer delay (CTD)
5. Cell delay variation (CDV)
6. Cell loss ratio for CLP0 (CLR0)
7. Cell loss ratio for CLP0+1 (CLR0+1)
8. Cell rate margin (CRM)
9. Variance factor (VF)
It is interesting to see the set of traffic parameters is very
focused on ATM technology only, although the PNNI routing protocol
uses the link state algorithm that is also used by other non-ATM
routing protocols. Perhaps this is a good example for us to
speculate that a routing protocol that uses IP addressing, link-
state based but focus on optical networks may also be built easily
from scratch, with much more cleaner background, and more reflective
and interactive with the underlined technology.
Bernstein, G. [Page 48]
draft-ipo-optical-inter-domain-00.txt November 2001
5.4.6 General Protocol Properties
This section briefly discusses the general protocol properties of
the PNNI in the specific areas as in the following sections.
5.4.6.1 System Overhead
The memory required to support PNNI operation is a variable
depending on several factors including the following:
1. Whether the node is a PGL or not, and if is, how many
hierarchical levels above for a PGL node in a given peer
group, there would be a separate PNNI node instance (LGN)
operating on the same ATM switch platform. This scenario is
also recursive in nature. Additional PNNI nodes in the same
hardware platform will consume more memory and CPU resources.
2. Whether the node is a border node or not border node may also
consume more memory and CPU resources since it needs to
generate and maintain Uplinks among other things.
3. The number of nodes and links within each PNNI peer group.
4. The number of reachable addresses within a single PNNI peer
group as well as advertised from other peer groups.
5. The amount of the traffic engineering information within a PNNI
peer group.
6. The total number of hierarchical levels.
Most of the PNNI routing messages are re-transmittable driven by the
associated timers in the order of seconds or milli-seconds and the
timer values are configurable.
5.4.6.2 Network Resource Overhead
PNNI routing messages can either be carried in-band, out-band but
always in-fiber as deployed as in todays ATM networks. The
consumption of the network resources depend on factors including the
following:
1. The amount of the routing messages that need to be put on the
wire - depends on factors that similar to those listed above
for the memory consumption.
2. The amount of the Resource Availability information for the
advertising. The code point for carrying Resource Availability
is defined in such a way where if more than one service
category with the same set of Resource Availability
information, a compact format can be used. E.g., if the set of
Bernstein, G. [Page 49]
draft-ipo-optical-inter-domain-00.txt November 2001
traffic parameters for CBR and rt-VBR on a link is the same,
there requires only one data block as defined by the code point
for both services, with a one-bit flag set for the indication.
In order to reduce network resource overhead, updates of
resource availability in PNNI is controlled by timers and
thresholds (defining significant change of a resource).
Optical network lack cell-specific resources described in
2.4.5., which should reduce the amount of advertised traffic.
3. The amount of the reachable addresses and address prefixes for
the advertising. If there exists a large number of reachable
addresses or address prefixes, they might be summarized before
the advertising. Note PNNI provides configurable summary
address that can be used for the address summarization.
PNNI routing protocol defines bandwidth and service category that
used for routing messages on RCCs. Default values are recommended
and also both are configurable as defined in the PNNI routing MIB.
This is a very good feature that needs to be taken into account for
the routing protocol used in the optical networks. One of the
reasons is that control channel used to carry routing messages in
the optical networks may be in-band, out-band or out-of-fiber, and
as such, the bandwidth consumption and perhaps other QoS
requirements for the routing messages might need to be estimated and
defined for reference.
5.4.6.3 Reliability
PNNI does not have re-start capability originally other than re-
build its routing database from scratch upon re-start. This fact
cannot be accepted by most of the carrier-class networks today, and
as a result, vendors will most likely to implement their own schemes
to support high reliability requirements. But the routing is not
service impacting unlike the call processing.
5.4.7 Scaling Capability
PNNI routing protocol is capable of scaling and can handle very
large ATM networks with the following characteristics:
-The scalability of PNNI routing protocol is the most comparing
to any other existing routing protocols due to its hierarchical
architecture. PNNI networks can scale up to 104 hierarchical
levels and at each level, there may be multiple peer groups.
Each PNNI peer group has up to from several 10s to several
hundred nodes in reality today.
-Like other routing protocols, PNNI achieves scaling capability
by partitioning networks into segments (peer groups in PNNI
case) and restrict the information flow across segment
Bernstein, G. [Page 50]
draft-ipo-optical-inter-domain-00.txt November 2001
boundary. The amount of the flow that floods into foreign PNNI
peer groups is controllable using protocol mechanism along with
configurable parameters.
-PNNI routing protocol is the only one as today can achieve
link state algorithm based routing both at the lowest level
hierarchy as well as all the higher level hierarchies. This
feature helps the effective routing across peer group
boundaries. Note link state algorithm based routing only exists
at the lowest level for both OSPF and IS-IS, not beyond.
The highlights of PNNI routing protocols scaling mechanism are as
follows:
1. Address summarization as described in the Section 2.4.3, PNNI
is capable of performing address summarization at each node of
each hierarchical level.
2. Link aggregation this is a similar feature as the link
bundling defined in the GMPLS but with much elegant
infrastructure. A set of horizontal links between two PNNI LGNs
may be aggregated into any smaller number of links by
configuration so-called Aggregation Token on the Outside Links
at the lowest hierarchical level. Upon aggregation, the amount
of Resource Availability information is also aggregated and
reduced.
3. Nodal aggregation this is a unique feature that no other
routing protocol possesses as today. A PNNI peer group is
represented by a LGN at the next higher hierarchical level and
as a default, the topology information inside the child peer
group is totally hidden from other peer group, exactly the same
scenario as that in an OSPF area. Therefore, the routing path
across PNNI peer group boundary has to be blindly by default
sent to one of the Exit Border node of the child peer group,
resulting the inefficient or sub-optimized routing paths. A LGN
that does not advertise any topology information for its child
peer group is called Simple Node Representation.Such a feature
properly extended could come in handy in the optical arena.
For example, the representations of a optical ring would be
quite simple (and this is a very popular optical subnetwork..
Optionally, a PNNI LGN can advertise some topology information for
the child peer group it represents, and such a LGN is called Complex
Node Representation. Usually a Complex node advertises a set of so-
called Nodal State PTSEs where each of them describes the set of the
traffic parameters (aggregated) between a pair of border nodes in
the associated child peer group; the information can be used during
Bernstein, G. [Page 51]
draft-ipo-optical-inter-domain-00.txt November 2001
the route calculation when choosing path traversing that peer group.
The PNNI routing protocol does not specify the limit of those
advertisements and how the aggregation is accomplished and this
provides flexibility for vendors own implementation as well as
customers preferences based on applications. The Complex Node
Representation provides a powerful and flexible solution to a common
topology scaling problem such that the routing efficiency and
optimization may also be possible across peer groups. PNNI Complex
Node representation is a logical star network with several
exceptions. As pointed out in the previous paragrpaph, for optical
networks, a different Complex Node representation may be needed
(perhaps, a logical ring with several exceptions).
The routing protocol used for the optical networks certainly
requires address summarization capability. It also certainly
requires the link aggregation capability since parallel links are
very common in optical networks. The need for nodal aggregation will
depend on the prediction of the size of the optical networks, i.e.,
it may not be needed the scope of a single PNNI peer group or a
single OSPF area can cover an entire given optical network. The
question is is this a correct assumption? Nodal aggregation at the
peer group level may be useful for dealing with some types of
"generalized node" diverse paths. But I'm not really sure, link
aggregation and the hierarchy stuff give us a lot already.
5.4.8 References
[PNNI1] ATM Forum, Private Network to Network Interface, af-pnni-
0055.000
[PNNI2] ATM Forum, I-LMI - Local Management Interface, af-ilmi-
0065.500
[PNNI3] ATM Forum, ATM Interim Inter-Switch Signaling Protocol,
af-pnni-0026.000
[PNNI4] ATM Forum, ATM Inter-network Interface (AINI), af-cs-0125-
00
6 Conclusion
This draft highlighted some of the considerations for an inter-
domain route protocol for use in optical internetworking. The main
differences between optical routing and datagram routing were
highlighted. Additional requirements to be addressed in an optical
inter-domain route protocol were discussed and several applications
of inter-domain routing were highlighted. A summary of optical
sublayer specific routing information was furnished for both the
transparent optical sublayer and the SONET/SDH sublayer. Finally a
review of the applicability of several existing route protocols to
the optical inter-domain route problem was given.
7 Security Considerations
Bernstein, G. [Page 52]
draft-ipo-optical-inter-domain-00.txt November 2001
7.1.1
Protection of the routing information exchanged across an optical
inter-domain interface is of high importance; erroneous reachability
or topology information may result in connection provisioning
requests that either fail or are routed across sub-optimal paths. It
is also possible that failed requests may consume significant
control and transport resources for a transient amount of time. It
follows that erroneous routing information could result in degraded
carrier network operation, or even render a carriers network
inoperable. Security requirements are expected to be of higher
importance in interfaces between different administrative domains.
Therefore, an optical inter-domain routing protocol should provide
the following:
1. Authenticate entities with which routing information is exchanged.
For example, a carrier should authenticate the identity of other
carriers it is connected to. The specific mechanisms used for
authentication should provide protection against attacks; for
example they should not be based on simple clear-text password
authentication schemes.
2. Guarantee the integrity of routing information (topology,
reachability and resource status) exchanged across the interface.
This requirement can be satisfied using security mechanisms at
different layers. For example, each routing message could be
individually authenticated using a keyed message digest, which is
embedded in the message. Both OSPF and BGP provide such options.
Alternatively, the two parties could establish a security
association at the network layer using IPSEC, which could be used
to provide security services to the optical inter-domain routing
protocol.
From the point of view of routing, information integrity is likely
to be the most important requirement. However, in some cases it
might be necessary to provide confidentiality of the routing
information as well. A possible scenario for this is when a carrier
would like to advertise information privately to another carrier,
but does not wish to publicly disseminate this information, due to
policy constraints.
It should be noted than none of the known mechanisms that provide
information integrity (such as keyed digests or IPSEC) can provide
adequate protection against a compromised node participating in the
inter-domain routing protocol. This is an item for further study.
8 References
[1] Bradner, S., "The Internet Standards Process -- Revision 3", BCP
9, RFC 2026, October 1996.
Bernstein, G. [Page 53]
draft-ipo-optical-inter-domain-00.txt November 2001
[2] G. Bernstein, J. Yates, D. Saha, "IP-Centric Control and
Management of Optical Transport Networks", IEEE Communications
Magazine, October 2000.
[3] G. Bernstein, E. Mannie, V. Sharma, "Framework for MPLS-based
Control of Optical SDH/SONET Networks", <draft-bms-optical-
sdhsonet-mpls-control-frmwrk-01.txt>, July 2001.
[4] Ramesh Bhandari, Survivable Networks: Algorithms for Diverse
Routing, Kluwer Academic Publishers,1999.
[5] Strand, J. (ed.) "Impairments And Other Constraints On Optical
Layer Routing", work in progress, draft-ietf-ipo-impairments-
00.txt, May 2001.
[6] Kompella, K., et. al. "IS-IS Extensions in Support of
Generalized MPLS", Work in Progress, draft-ietf-isis-gmpls-
extensions-01.txt, November 2000.
[7] K. Kompella, et. al. "Link Bundling in MPLS Traffic
Engineering", Work in Progress, draft-kompella-mpls-bundle-
05.txt, February 2001.
[8] ANSI T1.105.01-1995, Synchronous Optical Network (SONET)
Automatic Protection Switching, American National Standards
institute.
[9] Robert S. Cahn, Wide Area Network Design: Concepts and Tools
for Optimization, Morgan Kaufmann Publishers, Inc., 1998.
[10] Meghan Fuller, "Bandwidth trading no longer a case of 'if' but
'when' says report", Lightwave, June 2001. (www.light-wave.com)
[11] Awduche, D., et. Al., "RSVP-TE: Extensions to RSVP for LSP
Tunnels", Work in Progress, draft-ietf-mpls-rsvp-lsp-tunnel-
08.txt, February 2001.
[12] K. Owens, V. Sharma, M. Oommen, "Network Survivability
Considerations for Traffic Engineered IP Networks", Work in
Progress, draft-owens-te-network-survivability-01.txt, July 2001.
[13] K. Kompella and Y. Rekhter, "LSP Hierarchy with MPLS TE",
draft-ietf-mpls-lsp-hierarchy-02.txt, Internet Draft, Work in
Progress, February 2001.
9 Acknowledgments
Bernstein, G. [Page 54]
draft-ipo-optical-inter-domain-00.txt November 2001
The authors would like to thank Shezad Mirza of BT and Cheryl
Sanchez of Calix for their help in the routing protocol analysis
sections.
10 Author's Addresses
Greg Bernstein, Lyndon Ong
Ciena Corporation
10480 Ridgeview Court
Cupertino, CA 94014
Phone: (510) 573-2237
Email: greg@ciena.com, lyong@ciena.com
Bala Rajagopalan, Dimitrios Pendarakis
Tellium, Inc
2 Crescent Place
Ocean Port, NJ 07757
Email: braja@tellium.com, dpendarakis@Tellium.com
John Strand
AT&T Labs
200 Laurel Ave., Rm A5-1D06
Middletown, NJ 07748
Phone:(732) 420-9036
Email: jls@research.att.com
Angela Chiu
Celion Networks
1 Shiela Dr., Suite 2
Tinton Falls, NJ 07724
Phone:(732) 747-9987
Email: angela.chiu@celion.com
Frank Hujber
Alphion Corporation
4 Industrial Way West
Eatontown, NJ 07724
fhujber@alphion.com
Vishal Sharma
Metanoia, Inc.
335 Elan Village Lane, Unit 203
San Jose, CA 95134
Phone: +1 408 943 1794
Email: v.sharma@ieee.org
Sudheer Dharanikota
Nayna Networks Inc.
475 Sycamore drive,
Milpitas, CA 95035
Email : sudheer@nayna.com
Bernstein, G. [Page 55]
draft-ipo-optical-inter-domain-00.txt November 2001
Rauf Izmailov
NEC USA, Inc.
4 Independence Way
Princeton, NJ 08540
Email: rauf@nec-lab.com
Bernstein, G. [Page 56]