MPLS Working Group B. Jamoussi
Internet Draft Nortel Ltd
Expiration Date: February 1999
N. Feldman
IBM Corp
L. Andersson
Bay Networks Inc
August 1998
MPLS Ships in the Night Operation with ATM
<draft-jamoussi-mpls-sin-00.txt>
Status of this Memo
This document is an Internet-Draft. 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."
To learn the current status of any Internet-Draft, please check the
"1id-abstracts.txt" listing contained in the Internet-Drafts Shadow
Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe),
munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or
ftp.isi.edu (US West Coast).
Abstract
Multi-Protocol Label Switching (MPLS) can have several modes of
operation over ATM. The MPLS Framework document [1] indicates that
MPLS MUST allow 'ships in the night' (SIN) operation with existing L2
switching protocols (e.g., ATM Forum Signaling).
This document identifies the technical requirements that have to be
resolved in order to allow for a successful SIN operation between
MPLS and ATM Forum protocol stack. Solutions to the various
challenges are proposed.
Table of Contents
1 Introduction ......................................... 2
Jamoussi, et. al. August 6, 1998 [Page 1]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
1.1 Modes of Operation of MPLS over ATM .................. 2
1.2 Benefits of Operating MPLS over ATM .................. 3
1.3 Ships in the Night Mode of Operation ................. 4
2 VPI.VCI Space Partitioning ......................... 4
3 Traffic management ................................... 5
3.1 Bandwidth Management ................................. 7
3.2 Bandwidth Reservation ................................ 8
3.3 Queuing & Scheduling ................................. 9
3.4 Alignment with DS-Byte ............................... 10
4 Processing Capacity .................................. 11
5 Summary .............................................. 11
6 Security Considerations .............................. 11
7 Acknowledgment ....................................... 11
8 References ........................................... 11
9 Author's Address ..................................... 12
Appendix A Example of Equivalent Rate Computation ............... 13
Appendix B Example of ATM and MPLS COS mappings ................. 14
1. Introduction
In Section 1.1 we summarize the current models of using MPLS over
ATM. Section 1.2 highlights the benefits of running MPLS on ATM. In
Section 1.3, the requirements for running in Ships in the night (or
hybrid) mode are outlined.
1.1 Modes of Operation of MPLS over ATM
MPLS can have several modes of operation over ATM. This section
outlines the various operation modes that have been described so far
in various MPLS Internet Drafts. We classify MPLS over ATM modes of
operation as follows:
1. Use of ATM hardware as a Label-Controlled Interface:
a. MPLS-Only:
This mode has been described the most in the current MPLS
documentation. A label switching control component is
developed to control the ATM switching hardware as
described in [2].
b. Ships in the night:
This mode of operation has not been described in any level of
detail in any of the current MPLS drafts. The only mention
was to divide the VPI.VCI space between the ATM control plane
and the MPLS control plane [1, 2]. Other SIN operation
considerations were out of scope in [2].
Jamoussi, et. al. August 6, 1998 [Page 2]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
2. Use of ATM as a bearer service (B-ISDN):
a. Use of a VP:
A VP is established between two LSRs going across an ATM
network. The VCI is used as label within the VP [2, 3].
b. Use of a PVC or SVC and the VCID:
A PVC or an SVC is established across an ATM network.
The VCID is used to identify the two ends of the LSP [3, 4].
The choice of the model of deployment of MPLS on ATM depends on
various factors such as the pre-existence of an ATM network.
In the remainder of this document we focus on the ships in the night
mode of operation.
1.2 Benefits of Operating MPLS over ATM
Operating MPLS over an ATM network has many benefits described in
this section. Many of the ATM hardware features are readily available
to MPLS implementations.
- MPLS inherits the ATM hardware label (VPI/VCI) switching. This
allows an ATM-LSR to forward packets at very high rates.
- ATM's powerful traffic management features that are already
embedded in ATM hardware are available for use by MPLS with the
proper hooks in the MPLS control software (e.g., LDP). For example,
the queuing and scheduling techniques embedded in ATM hardware allow
it to provide many classes of service. This maps nicely to the COS
field in the LDP protocol. In addition, to queuing and scheduling,
you can also make use of traffic shaping and traffic policing at the
boundaries between MPLS domains.
- The use of ATM-LSRs allows for multiple services (voice, video, and
data) to be offered on a single platform. Operating MPLS over ATM
reduces the number of infrastructures required to provide multiple
services. It is hence possible to offer native ATM, a mixture of
Frame and ATM services, as well as MPLS using the same
infrastructure.
- Protect the capital investment in ATM infrastructure by increasing
the utilization of existing ATM network. Allow for an easy migration
from IP over ATM to IP over MPLS over ATM. Protect the network
management and operational knowledge.
Jamoussi, et. al. August 6, 1998 [Page 3]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
- While the MPLS architectures evolves, ATM offers traffic-
engineering solutions to solve today's bandwidth constraints.
1.3 Ships in the Night Mode
Ships in the night (SIN) operation is described in [1]. Section 1.2
of [1] includes the following requirement:
"MPLS MUST allow "ships in the night" operation with existing layer
2 switching protocols (e.g., ATM Forum Signaling) (i.e., MPLS must
be capable of being used in the same network which is also
simultaneously operating standard layer 2 protocols)."
SIN operation means that both MPLS and ATM control and routing
software are running on the same network. In addition, both MPLS and
ATM traffic share the same network infrastructure. However, both
protocols are oblivious to each other.
As the deployment of ATM networks continues to grow, it becomes
necessary for the successful deployment of MPLS to address the
question of how MPLS and ATM would co-exist.
The key advantage of running in SIN mode is that an existing Multi-
Service ATM network (running the ATM Forum software stack) can be
easily shared with MPLS (running IP routing and LDP). However, the
successful integration of MPLS and ATM on the same network requires
the following issues to be resolved:
1. VPI.VCI space partitioning
2. Traffic Management
3. Processing Capacity (Memory and CPU)
2. VPI.VCI Space Partitioning
In a label-controlled ATM (LC-ATM) interface, labels are the VPI.VCI
fields of the ATM cell. The entire 28-bit VPI.VCI field can be used
as a label in a flat hierarchy environment. However, a two level
hierarchy can also be achieved by using the VPI and the VCI fields
independently; each representing a level of the hierarchy.
ATM Forum stack makes use of the same VPI.VCI field to switch ATM
cells within a VP or a VC. Therefore, partitioning of the VPI/VCI
space (label space) between ATM and MPLS is necessary.
This issue is easily resolved through configuration. Two VPI.VCI
pools (Label-pool) are configured at start-up time. A Label-pool (a
sub-set of the VPI.VCI space) is allocated to ATM and another Label-
pool to MPLS. Each system only allocates VPI.VCI resources from its
Jamoussi, et. al. August 6, 1998 [Page 4]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
own Label-pool.
In keeping with the idea that both ATM and MPLS control planes are
oblivious to each other, it's necessary to avoid having one control
plane infer parameters based on the configuration of the other
control plane. Therefore, the division of label space between ATM and
MPLS needs to be pre-configured per interface. The exchange of the
valid VPI.VCI range between two adjacent LSRs/ATM switches is done as
follows:
- MPLS uses the LDP to negotiate the range of
valid VPI.VCI labels. This information is exchanged during the
initialization phase of the LDP session between two peer LSRs as
specified in LDP [5].
- ATM uses the ILMI channel to negotiate the valid VPI.VCI range as
specified in [6].
Both LDP and ILMI are currently defined to take the intersection of
the VPI.VCI range of two adjacent LSRs and ATM switches respectively.
The choice of the boundary between ATM and MPLS is a Network
Engineering exercise that takes into account the number of ATM VCs
that are expected, whether VC-merge is supported on the interface,
the level of stream granularity, among possibly other parameters.
3. Traffic Management
Traffic Management (TM) is one of the key components of MPLS.
Currently, TM is proposed to rely on explicit routing (ER), Class of
Service (COS) differentiation, and bandwidth reservation. [7]
provides an overall framework for TM in MPLS.
In [8], RSVP is used to setup ERs and reserve bandwidth. However, in
[5], the LDP is used to setup the ER and to reserve bandwidth for the
LSPs.
This Section defines the necessary Traffic Management building block
for providing CoS support in MPLS. In addition, it extends the
procedure defined in [5] by proposing a simple but effective Label
Switched Path Admission Control Mechanism that allows for high
statistical multiplexing gain while providing a bandwidth guarantee.
In order for MPLS to provide effective Class of Service (CoS)
differentiation and guarantees, a set of traffic management function
have to be implemented by the LSRs. These functions includes traffic
shaping, traffic policing, queuing and scheduling, and LSP admission
control.
Jamoussi, et. al. August 6, 1998 [Page 5]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
Traffic shaping is the TM function invoked at the egress point of a
network to limit the rate of traffic down to a specific value.
Traffic shapers often have a buffer that absorbs traffic bursts
beyond the shaping rate. Traffic shaping ensure that the outgoing
traffic rate from LSR1 to LSR2 (shown in Figure 1) does not exceed
the bandwidth agreement between ISP1 and ISP2. This function is
usually needed at the boundary between two administrative domains
(e.g., between ISP1 and ISP2).
| |
LSR1 | LSR2 LSR3 | LSR4
+------+ | +------+ +------+ | +------+
| | | | | | | | | |
| | | | | | | | | |
| S |---+-----| P A |---------| S |---+------| P |
| A | | | Q | | A | | | |
| Q | | | | | Q | | | |
| | | | | | | | | |
+------+ | +------+ +------+ | +------+
| |
ISP 1 | ISP 2 | ISP 3
| |
| |
Boundary Boundary
Legend
S Shaping
P Policing
A Admission Control
Q Queuing and Scheduling
Figure 1. MPLS Traffic Management Strategy
Traffic policing is the TM function invoked at the ingress point of a
network to ensures that only the agreed upon traffic rate is allowed
to consume network resources. Traffic policing is invoked by a
network service provider to protect its network resources (e.g.,
bandwidth) from traffic that exceeds the traffic contract. Traffic
policing ensure that the incoming traffic rate from LSR1 to LSR2 does
not exceed the bandwidth agreement between ISP1 and ISP2. This
function is usually needed at the boundary between two administrative
domains (e.g., between ISP1 and ISP2).
A queuing and scheduling system is required in order to provide
differentiated service levels. For instance three emission and four
discard priorities provide twelve different service levels. The
Jamoussi, et. al. August 6, 1998 [Page 6]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
emission priorities affect the delay observed by the traffic. The
discard priority affects the loss observed by the traffic. Delay
sensitive traffic uses the higher emission priority queue than less
delay sensitive traffic. Loss sensitive traffic uses the higher
discard priority (last one to discard) than the less loss sensitive
traffic. A sample queuing and scheduling system is described in
Appendix B.
Traffic shaping, policing, and queuing and scheduling are functions
often developed in hardware so that they don't impact data traffic
throughput. Most well-designed ATM hardware includes all of these
function. Therefore, an MPLS implementation on Label-Controlled ATM
interface (LC-ATM), inherits these functions.
Admission control ensures that bandwidth guarantees are met. This
function is often implemented in software and invoked at LSP setup
time. This Internet Draft proposes a simple yet effective mechanism
to reserve bandwidth for Explicitly Routed Label Switched Paths. We
focus in Sections 3.1-3.2 on LSP admission control which is the
procedure used to decide if a request for an LSP can be accepted,
based on the network capacity and the attributes of both the
requested and the existing LSPs. LSRs must perform LSPAC during the
LSP setup process as part of providing bandwidth guarantee.
3.1 Bandwidth Management
An important aspect of SIN operation is bandwidth management. ATM
and MPLS traffic share the same network facilities. The support of
SIN requires that the interaction between ATM traffic and MPLS
traffic is handled carefully.
A multi-service ATM network provides strict bandwidth guarantees to
its connections. MPLS Label Switched Paths (LSPs) can also reserve
bandwidth through the Label Distribution Protocol as defined in [5].
The bandwidth reservation issue can be resolved in various ways
through configuration at start-up time. The following configuration
options are possible:
- Hard partitioning between ATM and MPLS
The aggregate interface bandwidth is partitioned between
ATM and MPLS. Two bandwidth pools are created; one for ATM and
another for MPLS. Each pool is allocated a percentage of the
interface capacity. ATM makes its bandwidth reservations from the
ATM pool and MPLS makes its bandwidth reservations from the MPLS
pool.
The choice of the bandwidth pool boundary between ATM and MPLS
Jamoussi, et. al. August 6, 1998 [Page 7]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
is a Network Engineering exercise that takes into account the
expected bandwidth requirement of ATM and MPLS. An over-booking
factor can also be introduced where the sum of the percentages
allocated to ATM and MPLS add to more than 100%.
- Full sharing between ATM and MPLS
A single bandwidth pool is kept for both MPLS and ATM per
interface. As ATM connections or MPLS LSPs are admitted the
available bandwidth in the pool gets decremented.
The concept of bandwidth pools can be extended to provide a bandwidth
pool per ATM service category and per MPLS class of service in both
configuration modes.
3.2 Bandwidth Reservation in LDP
In [5], the following Reservation (RES) object is defined in Section
4.4.4.12:
"
+----------------+-------+--------------------------+----------+
| OBJECT | Type | Subtype(s) | Length |
+----------------+-------+--------------------------+----------+
| RES | 0x0C | 0x01 Raw Bandwidth | Variable |
+----------------+-------+--------------------------+----------+
SubType = 0x01 Raw Bandwidth
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| BW requirement |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
BW Requirement
Unsigned 32 bit integer representing the bandwidth, in units
of 64,000 bps, that must be reserved for the LSP at the LSR
identified in the ERNH Object that contains this Reservation
Object."
There is no clear indication on what the value of the BW Requirement
field in the RES object should be.
Using a similar terminology to the one used in [9], we define a
connection's traffic characteristics as follows:
- R: peak rate,
Jamoussi, et. al. August 6, 1998 [Page 8]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
- b: burst size,
- r: average rate
The BW requirement can be set to the peak rate R, average rate r,
or to an equivalent rate (e) of the LSP. Setting the BW requirement
to R of the LSP that has an On/Off traffic profile results in more
bandwidth reserved than actually needed. However, setting the BW
Requirement field to the r would result in high packet loss.
Therefore, in order to determine the BW Requirement field, it is
necessary to compute an equivalent rate of the Label Switched Path
(LSP). The computation of this rate needs to ensure two objectives
simultaneously:
1) Bandwidth guarantees; and
2) Efficient network resource utilization.
There are two ways of making the bandwidth reservation through LDP.
In the first option, the equivalent bandwidth is computed only once
at the LSR originating the ERLSP based on the traffic characteristics
(R, r, and b) of the LSP. The value e is then carried by LDP in the
RES object to inform the LSRs along the LSP of the bandwidth
requirement. The second option is to extend LDP to carry all three
traffic parameters and expect each LSR along the path to make its own
computation of the required bandwidth for the given LSP.
An extension to the current LDP specification [5] would be required
to support option 2 and to allow for the signaling of more bandwidth
parameters (R, r, and b) to more accurately characterize the
bandwidth requirements of an LSP.
Appendix A describes an example of an algorithm for computing e.
3.3 Queuing and Scheduling
ATM provides strict QoS guarantees. Hence, MPLS traffic should not
interfere with ATM traffic in any way that affect its QoS guarantees.
In addition, when MPLS provides bandwidth reservation and guarantees,
it is necessary to ensure that native ATM traffic does not interfere
with MPLS.
Bandwidth management presented in Section 3 works at the reservation
level. However, nodal-mechanisms of each LSR or ATM switch have to be
setup such that the QoS guarantees of ATM and MPLS are both met.
The nodal-mechanisms need to ensure that each service stays within
its bandwidth allocation limits. Traffic shaping and policing are
necessary to limit the throughput of each of the ATM and MPLS
Jamoussi, et. al. August 6, 1998 [Page 9]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
services to the contracted values.
In addition, loss and delay guarantees for both ATM and MPLS have to
be met. This requires a careful mapping of ATM service categories and
of MPLS Classes of Service on the shared queuing and scheduling
system.
Appendix B describes a Multi-Priority System (MPS) presented as an
example of a queuing and scheduling mechanism and the mapping of MPLS
COSs and ATM service categories to the MPS.
3.4 Alignment with the DS-byte
The Diff-Serv model indicates that traffic conditioning is done at
the edge of the network. A codepoint is assigned to define the
behavior aggregate of the packet. Within the core, packets are
forwarded based on the PHB associated with the DS codepoint. However,
when going over an MPLS network where the underlying infrastructure
is ATM, the DS codepoint is not accessible to the hardware making the
forward decision. Therefore, a mapping between the DS codepoint and
an MPLS LSP that would provide the same PHB is required as packets
enter the MPLS domain. This means that multiple LSPs are established
for a given FEC to accommodate the various DS codepoints.
Currently, only two PHBs are proposed; the DE and the EF PHBs. The
[9] draft indicates that the mapping from the DS-byte codepoint to
the behavior MUST be configurable. Since in an MPLS domain running on
ATM, the mapping from the DS-codepoint to the LSP is only feasible on
the edge of the MPLS domain, the mapping has to also be configurable.
The DS-byte includes an in/out bit that indicates whether the packet
is within its contracted rate or not. Setting this bit to 'out' means
that under congestion, this packet should be discarded before packet
that have the bit set to 'in'. The in/out bit in the DS-byte is
similar to the CLP bit in ATM cells. Therefore, as part of the
mapping function from the DS-byte codepoint to an LSP with a given
priority, the In/Out bit should be mapped to the CLP bit of all the
cells resulting from the segmentation of the frame.
When going from an MPLS-ATM domain to a Diff-Serv domain, a mapping
is also required from the various LSPs to a set of codepoints. In
addition, if the CLP bit of any cells of a frame are set, then the
"Out" bit needs to be set accordingly.
4. Processing Capacity
When running in SIN mode, two routing and control stacks will reside
on the same nodes. ATM requires its signaling, and routing (IISP or
Jamoussi, et. al. August 6, 1998 [Page 10]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
PNNI) software and topology database. MPLS requires its signaling
(LDP) and IP routing (RIP, OSPF, or BGP) software and topology
database.
Therefore, SIN operation requires that nodal processing capacity
(CPU, and memory) is adequate to simultaneously handle ATM and MPLS.
5. Summary
Ships in the night operation support is a required element in MPLS
[1]. This document highlights the requirements for SIN operation in
MPLS. Solutions are proposed to carefully share various resources
including VPI.VCI space, bandwidth, queuing and scheduling, and
processing capacity.
The traffic management building blocks that are necessary in order
for MPLS to provide CoS differentiation and guarantees are
identified. It proposed a simple yet effective mechanism to compute
the Equivalent Rate (e) of an Explicitly Routed Label Switched Path
to be signaled in the LDP RES object. The e computation allows for
high network bandwidth utilization while minimizing packet loss. An
admission control mechanism is described based on the e computation.
6. Security Considerations
Security issues are not discussed in this draft.
7. Acknowledgement
The authors would like to acknowledge the valuable comments of Pasi
Vaananen, Osama Aboul-Magd, Ken Hayward, and Don Fedyk. In addition,
recent MPLS Working Group discussions helped shape some sections of
this draft.
8. References
[1] R. Callon, et. al., "A Framework for Multiprotocol Label
Switching", draft-ietf-mpls-framework-02.txt, November 21, 1997.
[2] B. Davie, et. al., "Use of Label Switching With ATM", draft-
davie-mpls-atm-01.txt, July 1998.
[3] Ken-ichi Nagami, et. al., "VCID Notification over ATM link",
<draft-ietf-mpls-vcid-atm-00.txt>, March 1998.
[4] M. Suzuki, "The Assignment of the Information Field and Protocol
Identifier in the Q.2941 Generic Identifier and Q.2957 User-to-user
Signaling for the Internet Protocol", <draft-ietf-mpls-git-uus-
Jamoussi, et. al. August 6, 1998 [Page 11]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
00.txt>, June 1998.
[5] Anderson, et. al., "Label Distribution Protocol", draft-mpls-
ldp-00.txt, March 1998.
[6] "Integrated Local Management Interface (ILMI) Specification
Version 4", The ATM Forum technical committee, af-ilmi-0065.000, July
1996.
[7] P. Vaananen, et. al., "Framework for Traffic Management in MPLS
Networks", <draft-vaananen-mpls-tm-framework-00.txt>, March 1998.
[8] B. Davie, et. al., "Use of Label Switching With RSVP", <draft-
ietf-mpls-rsvp-00.txt>, March 1998.
[9] Y. Bernet, et. al., "A Framework for Differentiated Services",
<draft-ietf-diffserv-framework-00.txt>, May 1998.
9. Authors' Addresses
Bilel Jamoussi
Nortel (Northern Telecom), Ltd.
PO Box 3511 Station C
Ottawa ON K1Y 4H7
Canada
EMail: jamoussi@nortel.ca
Nancy Feldman
IBM Corp.
17 Skyline Drive
Hawthorne NY 10532
Phone: 914-784-3254
EMail: nkf@us.ibm.com
Loa Andersson
Bay Networks Inc.
3 Federal Street
Billerica, MA 01821
EMail: andersson@baynetworks.com
Jamoussi, et. al. August 6, 1998 [Page 12]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
Appendix A Example Equivalent Rate Computation
The LSP admission problem is to find a nominal rate, called the
equivalent rate (e), for each connection so that
the system meets the specified performance objectives.
This condition would be achieved as long as the sum of the e
values of the accepted LSPs does not exceed the capacity of
the designated link.
If the number of admitted LSPs exceeds the number indicated by
the bound, then the packet loss will likely exceed the required
target. However, if the number of admitted calls is less than this bound,
then link bandwidth is wasted.
The estimate of e is computed by taking the average of the minimum and
maximum number of LSPs that can be supported by the link.
The minimum number of LSPs, min, that can be admitted without
introducing any nodal loss or delay is obtained such that the
sum of the peak rates (R) is less than the link rate
(L) (min * R < L). Hence min=L/R.
The maximum number of connections, max,
is obtained when the sum of the average rates (r=A*R),
where A is the source activity (A=r/R), exceeds the
link rate (R * max < L), and there is sustained congestion.
Hence, max = L / (A*R). When dividing the link rate by the average
of min and max, we obtain the equivalent rate (e), given by (EQ1).
e = R * [(2*A) / (1+A)] (EQ 1)
Note that the link rate L simplifies in the EQ 1 and is no longer
needed in the computation of e.
After computing (EQ 1), only once per LSP, the resulting e is signaled
in the LDP message establishing the LSP. e is compared to the Available
Rate (AvR) value for each hop (link) as follows:
if (e < AvR)
accept the LSP;
else
reject the LSP;
Jamoussi, et. al. August 6, 1998 [Page 13]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
Appendix B Example of ATM Service Categories and MPLS COS Mappings
In this appendix, an example of how ATM and MPLS can be
mapped in a multi-priority system is presented.
Most well designed ATM hardware provides a way of differentiating the
quality of service received by the various ATM
service categories (CBR, VBR, UBR, etc.) through various queuing and
scheduling techniques.
In order to offer various delay and loss guarantees, ATM hardware often
uses multiple emission and discard priorities. Let's say we have 3
emission and 4 discard priorities. This MPS provides 12 different
qualities of service as shown in Figure 2. Traffic is emitted in the
order from E1 to E3. Therefore E1 has the lowest relative delay and E3
the highest. Under congestion, traffic is discard in the order from D4
to D1. Therefore, D1 has the lowest relative packet loss and D4 the
highest.
Emission
Priority
+-------------------------------------------+ ^
| | | CBR | Control | E1
| | | | |
+-------------------------------------------+
+-------------------------------------------+
| | CoS 2 | rt-VBR | Control | E2
| | | | |
+-------------------------------------------+
+-------------------------------------------+
| UBR | nrt-VBR | | Control | E3
| CoS 0 | CoS 1 | | |
+-------------------------------------------+
D4 D3 D2 D1
Discard Priority
Figure 2 Multi-Priority System (MPS)
Let's consider the open-loop ATM service categories (CBR, rt-VBR,
nrt-VBR, and UBR). For MPLS, we define three different Classes of
Service.
CoS 0 Best Effort
CoS 1 Low Loss Guarantee
CoS 2 Low Delay Guarantee
Jamoussi, et. al. August 6, 1998 [Page 14]
Internet Draft draft-jamoussi-mpls-sin-00.txt August 1998
Figure 2 also shows the mapping from the ATM service categories to the
MPS as well as the mapping of MPLS CoSs to MPS. CoS 0 and UBR have a
similar definition ("best effort"). Hence, they share the same emission
and discard priority (the lowest). CoS 1 and nrt-VBR are both targeted
for low loss guarantee but do not care too much about delay. Therefore,
they are mapped to the E3/D3 slot. CoS 2 and rt-VBR require Low delay
guarantee. Hence, they are mapped to a higher emission priority than
CoS 0/ CoS 1. In order to provide rt-VBR with slightly better loss
guarantee than CoS 2, rt-VBR is mapped to a higher discard priority
(D2). Finally, CBR requires the best low delay and low loss
guarantees. Hence it is mapped to the highest emission and discard
priority.
Network control traffic is in the case of ATM the signaling, ILMI, and
RCC channels and in the case of MPLS the LDP channel etc.
Jamoussi, et. al. August 6, 1998 [Page 15]