MEXT Working Group C. Larsson
Internet-Draft M. Eriksson
Intended status: Standards Track Ericsson Research
Expires: August 28, 2009 K. Mitsuya
Keio University
K. Tasaka
KDDI R&D Lab
R. Kuntz
Louis Pasteur University
February 24, 2009
Flow Distribution Rule Language for Multi-Access Nodes
draft-larsson-mext-flow-distribution-rules-02
Status of this Memo
This Internet-Draft is submitted to IETF in full conformance with the
provisions of BCP 78 and BCP 79.
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.
This Internet-Draft will expire on August 28, 2009.
Copyright Notice
Copyright (c) 2009 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
Larsson, et al. Expires August 28, 2009 [Page 1]
Internet-Draft Flow Distribution Rule Language February 2009
to this document.
Larsson, et al. Expires August 28, 2009 [Page 2]
Internet-Draft Flow Distribution Rule Language February 2009
Abstract
This document defines an OS independent rule language as a mean to
define and perform per flow path selection for a multi-homed node.
Per flow path selection is typically needed when there exist multiple
network interfaces, each with different network characteristics, and
an application has specific performance requirements for a data flow
that makes one network interface more suitable than another.
The flow distribution rule set is used by the node itself but also
exchanged with other nodes that needs to know about the multi-homed
node's capability of receiving data on multiple network interfaces.
This document does not define how the rule set is transferred between
nodes.
Larsson, et al. Expires August 28, 2009 [Page 3]
Internet-Draft Flow Distribution Rule Language February 2009
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1. Applicability . . . . . . . . . . . . . . . . . . . . . 5
1.2. Terminology . . . . . . . . . . . . . . . . . . . . . . 6
2. Architecture Overview . . . . . . . . . . . . . . . . . . . . 9
3. Conceptual Model . . . . . . . . . . . . . . . . . . . . . . . 11
3.1. Rule Set Characteristics . . . . . . . . . . . . . . . . 12
3.2. How to generate a PID . . . . . . . . . . . . . . . . . 14
3.3. Storing Routing Rules . . . . . . . . . . . . . . . . . 14
4. Multi-Access Rule Language Overview . . . . . . . . . . . . . 15
4.1. Rule Language Definition . . . . . . . . . . . . . . . . 15
4.2. Lexical Analysis . . . . . . . . . . . . . . . . . . . . 16
4.3. Rule Language Semantic . . . . . . . . . . . . . . . . . 16
5. Rule Set Semantics . . . . . . . . . . . . . . . . . . . . . . 17
5.1. Target Node . . . . . . . . . . . . . . . . . . . . . . 17
5.2. Rules and Path Identifier . . . . . . . . . . . . . . . 17
5.3. Conditional Rule-Sets . . . . . . . . . . . . . . . . . 17
5.4. Local and Peer Node . . . . . . . . . . . . . . . . . . 18
5.5. Any-port . . . . . . . . . . . . . . . . . . . . . . . . 18
5.6. Ranges . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.7. IPsec . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.8. ICMP . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.9. Explicit IP Protocol Numbers . . . . . . . . . . . . . . 19
5.10. Any IP Protocol and Flow Labels . . . . . . . . . . . . 20
5.11. Extra clauses . . . . . . . . . . . . . . . . . . . . . 20
5.12. Asymmetric Routing . . . . . . . . . . . . . . . . . . . 20
5.13. Round-robin . . . . . . . . . . . . . . . . . . . . . . 21
5.14. n-casting . . . . . . . . . . . . . . . . . . . . . . . 21
5.15. Mobile Networks . . . . . . . . . . . . . . . . . . . . 21
6. Generating Routing Rules and Bindings . . . . . . . . . . . . 22
6.1. Generating Routing Rules . . . . . . . . . . . . . . . . 22
6.2. Generating Bindings . . . . . . . . . . . . . . . . . . 22
6.3. Sending Routing Rules and Bindings to Peering Nodes . . 23
7. Security Considerations . . . . . . . . . . . . . . . . . . . 24
8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 25
9. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 26
10. References . . . . . . . . . . . . . . . . . . . . . . . . . . 27
10.1. Normative References . . . . . . . . . . . . . . . . . . 27
10.2. Informative References . . . . . . . . . . . . . . . . . 27
Appendix A. Applying the Rule Set to Mobile IPv6 . . . . . . . . 28
A.1. Mapping Between PID and IP Address . . . . . . . . . . . 28
A.2. Mobile IPv6 Example . . . . . . . . . . . . . . . . . . 28
Appendix B. Example of Routing Rules . . . . . . . . . . . . . . 32
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 34
Larsson, et al. Expires August 28, 2009 [Page 4]
Internet-Draft Flow Distribution Rule Language February 2009
1. Introduction
When a node is equipped with multiple network interfaces or has
multiple addresses assigned to one network interface, the node is
said to be multi-homed. When a multi-homed node establishes a
session with a peer, there exist several potential communication
paths between the nodes. Multiple communication paths between two
nodes imply that unless all traffic is sent over all links, there
must exist rules in the multi-homed node that for each packet
determine which path should be used to transmit the packet.
This document defines an OS independent rule language as a mean to
define and perform per flow path selection for a multi-homed node.
Per flow path selection is typically needed when there exist multiple
network interfaces, each with different network characteristics, and
an application has specific performance requirements for a data flow
that makes one network interface more suitable than another.
The rule language defined in this document is primarily used by
multi-homed nodes to describe a set of rules used for per flow path
selection. The rule set is also exchanged with other nodes (peers
and mobility anchor points) that needs to know about the multi-homed
node's capability of receiving data on multiple network interfaces.
The transfer of the rule set to the communicating peers is outside
the scope of this document.
Recently, some efforts have been dedicated to define IPv4-to-IPv6
transition mechanisms. It is most likely that multi-homed nodes will
connect to heterogeneous or dual-stack IP networks, hence the
motivation to support both versions of the IP protocol in the
definition of the rule language.
1.1. Applicability
The rule language defined in this document can be applied to both
stationary and mobile multi-homed nodes. The rule language is
agnostic with respect to the particular mobility management mechanism
used, and could therefore be used for any mobility management
protocols, e.g., Mobile IPv6 [RFC3775], Mobile IPv4 [RFC3344], HIP
[RFC5201], and MOBIKE [RFC4555].
The primary use of the defined rules, as described in this document,
is to specify to a peer node and a mobility anchor point the separate
communication paths to be used for multiple flows which pass through
the mobility anchor point (such as for instance the HA in a Mobile-IP
context) or originate at the peer, where the different flows have
different requirements for bandwidth, latency, and QoS (quality of
service). We will call the node which the multi-homed node decides
Larsson, et al. Expires August 28, 2009 [Page 5]
Internet-Draft Flow Distribution Rule Language February 2009
to exchange traffic with the 'Peer'.
Another example where using the defined rules may be useful is for a
moving multi-homed HIP-enabled node, when it has many sessions with
different QoS requirements towards the same server. The term 'Peer'
would refer to the server in this context.
To identify a mobile node some kind of identity is used. Some
examples of an identity is the home address in Mobile IPv6 [RFC3775]
and the Host Identity Tag (HIT) in HIP [RFC5201]. We will use the
term 'Identity Tag' as a name for this node identity. Additionally,
the multi-homed node would have local interface addresses associated
with each network interface. The binding between the identity tag
and the local interface addresses is handled by mechanisms specific
to each mobility management protocol (e.g., Mobile IPv6, HIP).
1.2. Terminology
This document frequently uses the following terms:
Binding
A binding is generally expressed in terms of a relation
between a Path Identifier (PID) and a local address. In
the context of multi-homed nodes, it is advantageous to
define match functions and bindings separately to avoid
excessive overhead, as it is expected that it may be
necessary to update the bindings much more often than the
match functions. Bindings can be updated on a handover to
a new local address, while the match function does not need
to change.
Identity Tag
The identity tag is the identity at which the multi-homed
node is addressable. Some examples of an identity tag are
the home address in Mobile IPv6 [RFC3775] and the Host
Identity Tag (HIT) in HIP [RFC5201].
Routing Proxy
A Routing Proxy is a node in the Internet that does routing
for the target node and needs to know about the multi-homed
node's capability of receiving data on multiple network
interfaces. For mobility protocols, this is the Anchor
Point, e.g., in case of Mobile IPv6 the Routing Proxy is
the Home Agent.
Larsson, et al. Expires August 28, 2009 [Page 6]
Internet-Draft Flow Distribution Rule Language February 2009
Local Node
In the context of multi-homed nodes, 'local node' refers to
the node for which the routing rules is aimed for. E.g.,
in case of Mobile IPv6 the local node is the Mobile Node.
Path Identifier (PID)
A Path Identifier (PID) identifies a path between a multi-
homed node and its peers. The PID maps to an interface on
the multi-homed node, how this is done depends on the
mobility mechanism. The PID is defined for the multi-homed
node, and its uniqueness is guaranteed by associating it
with the multi-homed node's identity tag. This procedure
will ensure that PIDs sent to a peer from different multi-
homed nodes will not collide.
Peer Node
A peer node is any node in the Internet that the multi-
homed node decides to exchange traffic with. E.g., in case
of Mobile IPv6 the peer node is the correspondent node.
Policy
In the context of multi-homed nodes, Policy is overall
information which express preferences and constraints on
how packets should be forwarded from the multi-homed node
to the intermediate node and the reverse path and from the
multi-homed node to the peer node and the reverse path.
Policy may cover such things as access network preferences,
user and operator preferences, security restrictions, and
more. Application of policy will in many cases result in
definition of routing rules which implement the policy for
specific traffic flows.
Routing Rule
A routing rule is a rule which specifies that traffic with
certain characteristics is to be handled in a certain
manner. As an example, a routing rule might specify that
any TCP traffic to or from port 80 should be transmitted on
a certain path.
A routing rule can be a selector and a PID. The selector
defines which packets match the routing rule. The PID
specifies the path, at a specific point in time, which
should be used for the matching packets.
Rule Set
A rule set is a collection of routing rules which is
associated with a certain decision point in the IP stack,
such as the point where a multi-access capable Mobile IP
Larsson, et al. Expires August 28, 2009 [Page 7]
Internet-Draft Flow Distribution Rule Language February 2009
implementation have to decide through which care-of address
a packet should be routed.
Larsson, et al. Expires August 28, 2009 [Page 8]
Internet-Draft Flow Distribution Rule Language February 2009
2. Architecture Overview
The term policy (or policies), in the context of multi-homed nodes,
refers to the overall settings and preferences that govern the
desired path selection between the multi-homed node and peer nodes.
The policies would typically include things such as the user's and
operator's preferences regarding access network technologies based on
certain characteristics, such as delay, bit error rate, cost of
usage, reliability, security, available bandwidth, etc. The
transmission and use of policies, to compute routing rules or for any
other use, is out of scope for this document.
The routing rules, in the context of multi-homed nodes, consists of a
selector and the path to use when a packet matches the selector.
This documents defines the rule language used for describing routing
rules.
Below are some examples of how routing rules would look like for
capturing traffic with certain characteristics and route it to
specific paths:
// Send HTTP traffic to peer using path 13
tcp peer port 80 on 13
// Use traffic class marking
udp tclass 127 on 14
any tclass 128-255 on 15
A peer node would typically be a node in the Internet that the multi-
homed node decides to exchange traffic with. E.g., in case of Mobile
IP the peer node is the correspondent node. The multi-homed node
sends routing rules to the peer nodes and intermediate nodes. An
example of an intermediate node in case of Mobile IP is the Home
Agent.
When an IP packet matches a routing rule, the result is to send it on
the path specified by the PID. When a PID has been associated with a
local address, and the local address is bound to a physical
interface, we have a complete specification of which physical
interface should be used to transmit a specific type of IP packet.
Figure 1 illustrates the conceptual separation for sending policies,
routing rules and binding information.
Larsson, et al. Expires August 28, 2009 [Page 9]
Internet-Draft Flow Distribution Rule Language February 2009
Local Node Peer Node
+-------------+ +-------------+
| | Exchange of Policies | |
| |<----------------------------->| |
| | | |
| | Exchange of Routing rules | |
| |<----------------------------->| |
| | | |
| | Binding PID <-> IP Address | |
| |<----------------------------->| |
+-------------+ +-------------+
Figure 1: Conceptual architecture overview
The proposed architecture of separating the exchange of policies,
routing rules and binding information is motivated by the fact that:
o The timing of events, which causes changes to the routing rules,
does not necessarily corresponds to a handover event and vice
versa.
o The routing rule exchange protocol can be used with any mobility
management protocol, e.g., MIPv6 [RFC3775], MIPv4 [RFC3344], HIP
[RFC5201] and MOBIKE [RFC4555].
Larsson, et al. Expires August 28, 2009 [Page 10]
Internet-Draft Flow Distribution Rule Language February 2009
3. Conceptual Model
Figure 2 is a conceptual model of how routing rules and bindings may
be generated. The Connection Manager may be implemented in any
manner consistent with the external behavior described in this
document.
+--------------+
Policies ---------> | |
| |
Events -----------> | | ---> Routing Rules
| Connection |
User/Operator ----> | Manager | ---> Bindings
preferences | |
| |
Access Network --> | |
Characteristics +--------------+
Figure 2: Conceptual model of how routing rules and bindings can be
generated.
In the context of this document a policy (or policies) is a high
level information which governs how traffic is sent from/to a multi-
homed node. One could think of policies as describing the preferred
actions that should be taken if certain conditions are fulfilled.
E.g., a multi-homed node could have a policy that states that if WLAN
is available then this interface is the preferred interface for
sending HTTP traffic. If WLAN is not available then the 3G interface
is the preferred interface for sending HTTP traffic.
A policy could be installed at a node prior to delivery and/or it
could be dynamically updated in run-time. Typically a node would
have a set of static policies installed while others are dynamically
installed when needed. The transmission, installation and use of
policies is outside the scope of this document.
Events could be anything that impact the current view of how
available resources should be used. For instance, when a new access
network becomes available this may cause new bindings to be created
for the existing set of routing rules. Events could also utilize the
current view to create routing rules and bindings. An example of
this would be when an application opens a socket, which would
typically generate new routing rules and bindings.
Preferences would be the user's and operator's way of impacting how
different access technologies are used. One way of controlling this
Larsson, et al. Expires August 28, 2009 [Page 11]
Internet-Draft Flow Distribution Rule Language February 2009
could be by the user's subscription. Subscriptions could be in the
range of "operator decides everything" to "user decides everything".
Each access network has certain characteristics, such as loss rate,
jitter, latency, bandwidth, QoS support, power consumptions etc.,
which impact the choice of access technology for a service. Some
access characteristics are static while other are dynamic, and the
changes could be viewed as events.
Policy, events, preferences and access network characteristics are
examples of input to the Connection Manager, which generates routing
rules and bindings based on this input. The specification of the
Connection Manager is outside the scope of this document.
The routing rules consists of a selector and the PID to use when an
IP packet matches the selector. This document defines the language
used for describing the routing rules. The association between a PID
and the actual IP address is called a Binding. The Connection
Manager application is responsible for the creation of bindings.
3.1. Rule Set Characteristics
A collection of routing rules is called a Rule Set, and refers to the
current mapping of flows on the local node's available access
networks. Typically, one common rule set is generated for the local
node and all the peer nodes which the local node is communicating
with. However, the routing rule syntax makes it possible to specify
a routing rule targeting a specific peer.
The multi-homed node or a trusted network node could generate the
routing rules. The routing rules are defining the routing for a
specific node (or a specific mobile network). Since the rule set is
common for the multi-homed node and its peering nodes, the role of
the node applying the routing rules is important when interpreting
the routing rules. The keyword 'local' means the multi-homed node
and the keyword 'peer' means the peering nodes.
Below is an example showing a rule set for a multi-access node (MA)
with two communication paths, one over interface I1 (identified by
PID1) and another one over interface I2 (identified by PID2),
connected to two different access networks. The multi-access node is
communicating with two peering nodes, P1 and P2, both being single-
homed.
Larsson, et al. Expires August 28, 2009 [Page 12]
Internet-Draft Flow Distribution Rule Language February 2009
__----__
( Access )
( Network ) ___----___ +----+
+------+ _ /(__ __) \ ( )--------| P1 |
| I1|_/ ---- \_( ) +----+
| MA | _( Internet )
| I2|_ __----__ / ( ) +----+
+------+ \__ ( Access )_/ (___ ___)--------| P2 |
( Network ) ---- +----+
(__ __)
----
The rule set on the multi-homed node depends on the type of
communication established with the peers. In this example the multi-
homed node has established HTTP, interactive SSH and peer-to-peer
voice over IP traffic with its peers (P1 and P2). The PIDs PID1 and
PID2 maps to I1 and I2, respectively:
tcp peer port 80 on PID1
tcp peer port 22 on PID2
udp local port 49724 peer P2 port 56512 on PID2
The above rule set is distributed to the peer nodes. How the rule
set is transferred to the peering nodes is outside the scope of this
document. Whether the whole rule set or a subset of the rule set is
sent to the peering nodes depends on the transport protocol in use
and if there is any need for the peer node to have this type of
knowledge for its communication with the multi-homed node. If, in
the above example, the MA node would have established a HTTP session
with P1 and a HTTP and VoIP session with P2, then P1 would not need
any routing rules to send the return traffic back to the MA.
However, P2 would need to know which communication path to use for
the different traffic flows.
The implementation of the actual routing mechanism to match the rule
selectors and follow the rules is platform dependent.
Larsson, et al. Expires August 28, 2009 [Page 13]
Internet-Draft Flow Distribution Rule Language February 2009
3.2. How to generate a PID
How the Connection Manager generates the PID is outside the scope of
this document. For example, one possible way of doing it would be
for the Connection Manager to assign a unique PID number for all
traffic requesting the same (or similar) type of service (e.g.,
bandwidth, bit error rate, latency, etc.). If the access conditions
changes, i.e., the Connection Manager receives new input data, new
bindings would be generated which could change the physical interface
that an existing PID number is associated with.
The PID is created by the multi-homed node and its uniqueness is
guaranteed by associating it with the multi-homed node's identity
tag.
3.3. Storing Routing Rules
How the routing rules are stored, in the mult-home node and in the
intermediate node and the peers, is implementation dependent. They
could be stored in a file system in the ASCII form that is described
in this document, but they could also be stored in, e.g., in-core
(binary) data structures.
Larsson, et al. Expires August 28, 2009 [Page 14]
Internet-Draft Flow Distribution Rule Language February 2009
4. Multi-Access Rule Language Overview
This section defines the syntax used to describe routing rules. It
uses the formal syntax Augmented Backus-Naur Form (ABNF) as specified
in [RFC5234].
4.1. Rule Language Definition
rule-collection = RULE-SET / 1*(CONDITIONAL-SET)
conditional-set = CONDITION RULE-SET
condition = "<" NUMLIST ">"
rule-set = *(RULE ';')
rule = [ AF ] FLOW EXTRA ACTION [ WHERE ]
af = "ip4" | "ip6"
flow = ("tcp" / "udp") (PORT-ADDR-PAIR / ANY-PORT)
flow =/ ("ipsec" / "ah" / "esp") SPI-ADDR-PAIR
flow =/ "icmp" [ ICMP-SPEC ] ADDR-PAIR
flow =/ "proto" NEG-RANGE ADDR-PAIR
flow =/ "any" LABEL-ADDR-PAIR
extra = [ "hoplimit" NEG-RANGE ] [ "tclass" NEG-RANGE ]
[ "ip6h" NEG-RANGE ]
extra =/ [ "tos" NEG-RANGE ] [ "ttl" TTL ]
action = "on" (RR-LIST / CAST-LIST) / "drop"
where = "at" ("local" / PREFIX)
port-addr-pair = [ "local" PORT-ADDR ] [ "peer" PORT-ADDR ]
port-addr = [ PREFIX ] PORT-SPEC / PREFIX
any-port = PORT-SPEC ADDR-PAIR
spi-addr-pair = [ "local" SPI-ADDR ] [ "peer" SPI-ADDR ]
spi-addr = [ PREFIX ] "spi" NEG-RANGE / PREFIX
label-addr-pair = [ "local" LABEL-ADDR ] [ "peer" LABEL-ADDR ]
label-addr = [ PREFIX ] "label" NEG-RANGE / PREFIX
addr-pair = [ "local" PREFIX ] [ "peer" PREFIX ]
icmp-spec = "type" NEG-RANGE [ "code" NEG-RANGE ]
rr-list = NUMLIST
cast-list = NUMBER 1*("&" NUMBER)
prefix = ADDR [ "/" NUMBER ]
addr = HEXSEQ [ "::" [ HEXSEQ ] ] / "::" [ HEXSEQ ]
addr =/ 1*3DIGIT 3("." 1*3DIGIT)
hexseq = HEX4 *(":" HEX4)
hex4 = 1*4HEXDIGIT
port-spec = "port" NEG-RANGE
neg-range = [ NOT ] RANGE
not = "!" / "not" / "no"
Larsson, et al. Expires August 28, 2009 [Page 15]
Internet-Draft Flow Distribution Rule Language February 2009
range = NUMBER [ "-" NUMBER ]
numlist = NUMBER *("," NUMBER)
number = DEC-NUMBER / HEX-NUMBER
dec-number = 1*DIGIT
hex-number = "0x" 1*HEXDIGIT
hexdigit = DIGIT / "A" / "B" / "C" / "D" / "E" / "F"
digit = %x30-39
4.2. Lexical Analysis
The routing rules are free text in ASCII text format. White space is
used for token separation. Allowed white space is Space (ASCII 32),
Horizontal Tab (ASCII 9), and CRLF (ASCII 13 10).
Comments are allowed at any place where white space is allowed, and
will be parsed as white space. A comment starts with two slashes
('/', ASCII 47), and continues to the next CRLF.
The routing rule language is not line based, but there is a 1024
octet limitation on the length of lines (including the final CRLF),
to simplify implementations in memory constrained devices.
4.3. Rule Language Semantic
The rule language definition presented in section 4.1 may lead to
semantic errors though the rule set is syntactically correct: for
example, the language definition allows to mix IPv6 addresses with
the Type of Service (tos) field in the rule selector, though the IPv6
header does not have such field.
Such design highly simplifies the language definition, but mandates
sanity checks when implementing a language parser.
Larsson, et al. Expires August 28, 2009 [Page 16]
Internet-Draft Flow Distribution Rule Language February 2009
5. Rule Set Semantics
This section explains how the syntax should be interpreted and the
motivation for using this specific syntax.
5.1. Target Node
The node for which the routing rules are used is called the target
node. The Identity Tag of the target node is given out-of-band, when
the rules are transmitted or installed. Exactly how this is done
depends on mobility mechanism used, and is outside the scope of this
document.
It is also possible to use the rules for a whole moving network, see
below.
5.2. Rules and Path Identifier
The routing rules will identify packet flows, and result in flows
being dropped or mapped to a particular Path Identifier, PID. How
the PID is then mapped to an actual packet transmission channel
depends on the mobility management method used, and is outside the
scope of this document.
When a packet is routed, the rules are checked sequentially. The
first rule that matches the packet is selected. What happens if no
rule matches, depends on the mobility mechanism, and is not defined
here. If the mobility mechanism supports it, it is suggested that
the packet will use a default path, otherwise the packet will
probably have to be dropped. It is strongly advised that all rule
sets have a catch-all default rule, and does not depend on the
behavior for non-matching packets.
5.3. Conditional Rule-Sets
For some scenarios, it is useful to install or transmit a collection
of rule-sets at the same time, and select which one to use depending
on what PIDs are available. How the set of available PIDs is
determined is outside the scope of this document.
When multiple rule-sets are put in a rule-collection, each rule set
is conditioned with a list of PIDs that should be available inside
angle brackets, '<' and '>'. An example:
Larsson, et al. Expires August 28, 2009 [Page 17]
Internet-Draft Flow Distribution Rule Language February 2009
<11,800>
tcp local port 80 on 800
any on 11
<11>
tcp local port 80 drop
any on 11
In the rule collection above, the first rule set will be used when
both PIDs 11 and 800 are available. The second rule set will be used
when only PID 11 is available. If no condition matches, an empty
rule set will be used, which implies that all packets are dropped.
5.4. Local and Peer Node
It is an explicit design goal of the language that the very same
rules can be used at all nodes that route traffic to or from the
target node. This implies that the flow description does not express
header source or destination, because they will have to be switched
depending on the direction of the packet transmission. To achieve
direction independence, the keyword 'local' and 'peer' are used to
refer to the endpoints of the flows. The target node is referred to
as 'local', and whatever node the target node is communicating with
is referred to as 'peer'. An example:
tcp peer port 80 on 11
This rule will send packets from the target node with destination
port 80 on PID 11, and correspondingly send packets to the target
node with source port 80 on the same PID. In practice, this means
that any TCP traffic on the standard HTTP port to or from web servers
will be put on PID 11.
5.5. Any-port
When using well-known ports, the port number can be used to identify
the type of traffic. The port number is normally checked at one of
the endpoints, 'local' or 'peer'. In some cases, it doesn't matter
which endpoint that uses the well-known port, the fact that either
one does is enough to classify the traffic. Traffic using a well-
known port could be handled by two rules together:
tcp local port 25 peer 2001:db8::1 on 44
tcp peer 2001:db8::1 port 25 on 44
The first rule would handle SMTP connections from any port on 2001:
db8::1 to the Mail Transfer Agent, MTA, at the target node. The
second rule would correspondingly handle SMTP connections from any
port on the target node to the MTA at 2001:db8::1. Note that the
Larsson, et al. Expires August 28, 2009 [Page 18]
Internet-Draft Flow Distribution Rule Language February 2009
rules can not be collapsed like this:
tcp local port 25 peer 2001:db8::1 port 25 on 44
This would only match flows using port number 25 at both ends, which
is normally not the case.
To avoid the double rules in the common case of traffic
classification on (well-known) port numbers, the any-port mechanism
can be used. When used, it means that at least one of the local and
peer port number must match. An example:
tcp port 25 peer 2001:db8::1 on 44
This rule would handle any SMTP traffic between the target node and
2001:db8::1, and has the same meaning as the two rules above have
together. Note that the port number is given before any 'local' or
'peer' clauses.
5.6. Ranges
At many places where a number is given, a NEG-RANGE can be used. It
is a numerical range, which can be optionally negated. Some examples
where ranges are used for port numbers:
tcp port 49152-65535 on 13
tcp port not 137-139 on 14
5.7. IPsec
IPsec traffic can be recognized as one of the IP protocols AH and
ESP. The keyword 'ipsec' is a shortcut for any of AH and ESP. When
matching IPsec traffic, the SPI can be used for more specific
matching:
esp local spi 1500-1599 on 11
esp peer 2001:db8::1 spi 444 on 11
ipsec on 44
5.8. ICMP
ICMP packets can be matched on type and code.
5.9. Explicit IP Protocol Numbers
It is possible to match flows on protocol numbers:
proto 138-255 drop
Larsson, et al. Expires August 28, 2009 [Page 19]
Internet-Draft Flow Distribution Rule Language February 2009
This will drop any packets with IP protocol number between 138 and
255.
5.10. Any IP Protocol and Flow Labels
If the IP protocol number should not be checked, the 'any' keyword
can be used. An 'any' rule allows matching on the IPv6 header flow
label:
any peer 2001:db8::1 flow 99 on 15
any local flow 42 on 15
any on 17
5.11. Extra clauses
It is possible to match each packet on a few other things:
o The IPv6 header hop limit
o The IPv6 header traffic class
o Whether some particular extension headers are included
o The IPv4 Type of Service field
One or more of these checks can be included in each rule.
5.12. Asymmetric Routing
Asymmetric routing means that the packets of a particular flow is not
using the same path in both directions. This can be useful when
using networks that are intrinsically asymmetric, like satellite
networks.
One way to achieve asymmetric routing is to have different rules at
the target node and at the routing proxy or peers. In some
scenarios, it is valuable to have the very same rules at all places
even in the case of asymmetric routing. To support this, each rule
can have an "at" clause. The "at" clause means that the rule is only
applicable at that address or prefix. As a special case, the keyword
'local' is used to refer to the target node. An example:
udp peer 2001:db8::15 on 11 at local
udp peer 2001:db8::15 on 22
The first rule will only be applied at the target node, so PID 11
will be used for UDP traffic from the target node to 2001:db8::15.
All other nodes (routing proxy or peer nodes) will apply the second
Larsson, et al. Expires August 28, 2009 [Page 20]
Internet-Draft Flow Distribution Rule Language February 2009
rule, and UDP traffic from 2001:db8::15 to the target node will use
PID 22.
5.13. Round-robin
For load-sharing purposes, round-robin transmission is supported. To
use round-robin, a list of PIDs is given instead of just one:
udp peer 2001:db8:1c32::/48 port 44100 on 4, 4, 5
Here, UDP traffic from port 44100 at any node in prefix 2001:db8:
1c32::/48 will be sent on PIDs 4 and 5. Two out of three packets
will go on PID 4, which presumably has twice the bandwidth of PID 5.
5.14. n-casting
In some specific cases, like important signaling, it can be useful to
send the packets on multiple PIDs in parallel. All type of traffic
is not suited for n-casting. For instance, TCP is known to be
sensitive for re-ordering of packets and n-casting may cause TCP slow
start. However, the routing rules provides a mechanism to handle
transport protocols differently, which would make it possible to use
n-casting for, e.g., some UDP traffic:
udp peer port 5060 on 11 & 800
This would transfer UDP packets to and from peer port 5060 on both
PID 11 and PID 800.
5.15. Mobile Networks
A set of rules can be used for an entire moving network. By default,
'local' will then refer to all the nodes of the mobile network. When
a particular MNN (or set of MNNs) needs special routing, an address
or prefix can be used after 'local'. An example:
udp local 2001:db8::19 port 5600 on 18
udp local port 5600 on 19
The first rule will give special treatment to the MNN 2001:db8::19,
all other MNNs will be handled by the second rule.
Larsson, et al. Expires August 28, 2009 [Page 21]
Internet-Draft Flow Distribution Rule Language February 2009
6. Generating Routing Rules and Bindings
This section describes when and how routing rules and bindings are
generated and sent from the local node to the intermediate and
peering nodes.
6.1. Generating Routing Rules
Routing rules could be generated either by the multi-homed node
itself or by a trusted network node on behalf of the multi-homed
node. The multi-homed node is an obvious candidate for generating
routing rules and bindings since it's aware of all the information
needed by the Connection Manager. It would also be able to react
instantly whenever an event occurs that requires an immediate action.
However, there may be situations when a trusted network node could
generate some or all routing rules and bindings. If everything is
generated by this node additional communication between the multi-
homed node and this trusted network node would be required. The
actual implementation of this is outside the scope of this document.
The frequency at which the routing rules are generated is an
implementation issue. Depending on the target node type, the
Connection Manager may generate the whole set of routing rules as the
node is initiated or it could prefer to have a more dynamic approach
and generate the routing rules on demand as the node initiate
applications.
In case of a static approach the required routing rules exists once
the multi-homed node has been initiated and it's assumed that they
will change with a low frequency.
In case of a dynamic approach there are several events that may
trigger the creation of a new routing rule. For example when an
application requests to open a socket the Connection Manager
application would create a new routing rule with an associated PID
number. As the user starts different applications, the rule set will
expand and there will be multiple routing rules, each of one
targeting a specific data flow, with information of how the multi-
homed node prefer inbound and outbound traffic to be routed. The
rule set is dynamic in the sense that if an application is terminated
then associated routing rules are removed.
6.2. Generating Bindings
The Binding is a mapping between the PID and a physical IP address.
How this mapping is done depends on the mobility mechanism in use and
is outside the scope of this document. The syntax defines a selector
which is associated with a PID.
Larsson, et al. Expires August 28, 2009 [Page 22]
Internet-Draft Flow Distribution Rule Language February 2009
There are several events that may trigger the creation of a new
binding. For example when an interface becomes available or is
deactivated, or when the signal strength goes below a certain
threshold value, etc. All these events are input to the Connection
Manager that evaluates it, based on certain criteria that have been
defined by the user and/or operator, and generates a new set of
bindings that reflects the current status of the multi-homed node's
communication capacity.
6.3. Sending Routing Rules and Bindings to Peering Nodes
When a new routing rule or a binding has been created, this
information must be sent to the intermediate and peering nodes to
make sure that these nodes knows which destination address to use if
there exist multiple addresses associated with the multi-homed node's
identity tag.
Whether the entire rule set or a subset of it is sent to a peering
node depends on the transport protocol in use and is outside the
scope of this document.
Larsson, et al. Expires August 28, 2009 [Page 23]
Internet-Draft Flow Distribution Rule Language February 2009
7. Security Considerations
This document specifies a language and not a protocol. Hence, there
are no inherent security issues related to this specification.
However, when the routing rules and bindings are distributed and
installed, authentication and authorization must be ensured. Exactly
how this is done is outside the scope of this document.
Additionally, the address scope of the routing rules (in particular
rules using "local <prefix>" clauses) must be checked, to not include
addresses outside the allowed range. Failure to do these checks can
make it possible to do unauthorized routing changes for network
traffic to other hosts.
Larsson, et al. Expires August 28, 2009 [Page 24]
Internet-Draft Flow Distribution Rule Language February 2009
8. IANA Considerations
This memo includes no request to IANA.
Larsson, et al. Expires August 28, 2009 [Page 25]
Internet-Draft Flow Distribution Rule Language February 2009
9. Acknowledgements
We would like to thank (in alphabetic order) Tero Kauppinen, Henrik
Levkowetz, Heikki Mahkonen, and Ryuji Wakikawa for their comments and
suggestions on this work.
Larsson, et al. Expires August 28, 2009 [Page 26]
Internet-Draft Flow Distribution Rule Language February 2009
10. References
10.1. Normative References
[RFC5234] Crocker, D. and P. Overell, "Augmented BNF for Syntax
Specifications: ABNF", STD 68, RFC 5234, January 2008.
10.2. Informative References
[I-D.draft-eriksson-mext-mipv6-routing-rules]
Eriksson, M., "Mobile IPv6 Flow Routing over Multiple
Care-of Addresses",
Internet-Document draft-eriksson-mext-mipv6-routing-rules,
June 2008.
[I-D.ietf-mext-flow-binding]
Soliman, H., Montavont, N., Fikouras, N., and K.
Kuladinithi, "Flow Bindings in Mobile IPv6 and Nemo Basic
Support", draft-ietf-mext-flow-binding-01 (work in
progress), February 2009.
[I-D.ietf-monami6-multiplecoa]
Wakikawa, R., Devarapalli, V., Ernst, T., and K. Nagami,
"Multiple Care-of Addresses Registration",
draft-ietf-monami6-multiplecoa-11 (work in progress),
January 2009.
[RFC3344] Perkins, C., "IP Mobility Support for IPv4", RFC 3344,
August 2002.
[RFC3775] Johnson, D., Perkins, C., and J. Arkko, "Mobility Support
in IPv6", RFC 3775, June 2004.
[RFC4555] Eronen, P., "IKEv2 Mobility and Multihoming Protocol
(MOBIKE)", RFC 4555, June 2006.
[RFC5201] Moskowitz, R., Nikander, P., Jokela, P., and T. Henderson,
"Host Identity Protocol", RFC 5201, April 2008.
Larsson, et al. Expires August 28, 2009 [Page 27]
Internet-Draft Flow Distribution Rule Language February 2009
Appendix A. Applying the Rule Set to Mobile IPv6
This appendix draws examples on how to use routing rules with Mobile
IPv6 [RFC3775]. Other mobility management protocols would need to
write separate documents to outline how the routing rules are used
with that specific mobility management protocol.
The multiple care-of registration protocol
[I-D.ietf-monami6-multiplecoa] defines a way to maintain multiple
paths between two nodes. However, both communicating nodes must have
routing rules to know how to distribute traffic flows between the
different paths. There may be different ways of distributing the
routing rules. In case of Mobile IPv6 this could be done as defined
in either [I-D.ietf-mext-flow-binding] or
[I-D.draft-eriksson-mext-mipv6-routing-rules].
A.1. Mapping Between PID and IP Address
In the conceptual model, the output from the Connection Manager is
Bindings. The binding is the actual mapping between the PID and the
physical IP address. By using a PID in the routing rules, a dynamic
binding between the PID in the routing rule and the actual physical
interface is achieved. In case of Mobile IPv6 the physical IP
address is a MIP care-of address.
The base Mobile IPv6 [RFC3775] specification only supports that one
Care-of Address (CoA) is bound to the Home Address (HoA). To achieve
simultaneous multi-access, the base protocol needs extensions and
document [I-D.ietf-monami6-multiplecoa] defines how to register
multiple care-of addresses bound to a single Home Address. For doing
so, a new Binding Unique Identification Number (BID) is carried in
each binding for the receiver to distinguish between the bindings
corresponding to the same Home Address.
The Connection Manager creates routing rules with the associated PID
and the binding between the PID, BID and care-of address as
illustrated below.
Routing Rule: PID -----> BID -----> CoA
In case of Mobile IPv6 the PID would typically be the BID, i.e.,
there is no need for an explicit additional mapping.
A.2. Mobile IPv6 Example
This example is an extension of the example given in Chapter 4 and
shows how it can be applied to Mobile IPv6. The Mobile Node (MN)
Larsson, et al. Expires August 28, 2009 [Page 28]
Internet-Draft Flow Distribution Rule Language February 2009
with two communication paths, one over interface I1 (identified by
PID1) and another one over interface I2 (identified by PID2),
connected to two different foreign access networks. The MN is
communicating with two Correspondent Nodes, CN1 and CN2, either
directly if Route Optimization is used or via the Home Agent (HA) if
reverse tunneling is used. The IPv6 address assigned to CN2 is 2001:
db8:1411::24. Both CN1 and CN2 are single-homed.
+----+
| HA |
+----+
__----__ |
( Access ) |
( Network ) ___----___ +-----+
+-------+ _ /(__ __) \ ( )--------| CN1 |
| I1|_/ ---- \_( ) +-----+
| MN | _( Internet )
| I2|_ __----__ / ( ) +-----+
+-------+ \__ ( Access )_/ (___ ___)--------| CN2 |
( Network ) ---- +-----+
(__ __)
----
In this example the Connection Manager receives different input
events, some of them causing new routing rules and bindings to be
created. We assume the following scenario:
o The MN connects interface I1 to a Wide Area Network (WAN) and
interface I2 to a Wireless LAN (WLAN) network. From the MN point
of view both access networks are foreign networks.
o As the MN connects to the access network this will trigger the
Connection Manager to generate a binding. For interface I1, BID1
is assigned number 800, and for interface I2, BID2 is assigned
number 11. In addition, BID1 is mapped to CoA1, which has been
assigned to interface I1 and BID2 is mapped to CoA2, which has
been assigned to interface I2. How this mapping is achieved is
outside the scope of this specification.
o According to [I-D.ietf-monami6-multiplecoa] the newly created
bindings are sent to the HA where a Binding Cache entry for the
MN's HoA and CoA1 and CoA2 is created.
o Now the user at the MN decides to establish a HTTP session towards
CN1 and at the same time call a friend that can be reached at CN2.
These events (applications open sockets) are input to the
Larsson, et al. Expires August 28, 2009 [Page 29]
Internet-Draft Flow Distribution Rule Language February 2009
Connection Manager and as a result a set of routing rules are
created that reflects the current view of the MN on how these
traffic flows should be routed. In this example the Connection
Manager has decided to put HTTP traffic on interface I2 and the
voice traffic on interface I1.
tcp peer port 80 on 11
udp local port 49724 peer 2001:db8:1411::24 port 56512 on 800
o These routing rules will ensure that outgoing traffic from the MN
is sent to the correct interface, but to also ensure that incoming
traffic towards the MN is routed through the correct interface the
HA also needs access to the MN's routing rules. Hence, the MN
will send the newly created routing rules to the HA. In case of
Mobile IPv6 this could be done as defined in either
[I-D.ietf-mext-flow-binding] or
[I-D.draft-eriksson-mext-mipv6-routing-rules].
o Whether the above routing rules also must be sent to the
correspondent nodes depends on if the MN will do route
optimization towards the CN and if the MN decides to register
multiple CoAs with its HoA. If this is the case, then the CN will
also need the routing information to able to make the correct
choice of CoA for the traffic sent towards the MN.
The routing rules in this example would be interpreted differently by
the MN and the HA/CN. In case of the MN it would "send HTTP traffic
with destination port 80 using CoA2 as source address" and "send
voice over IP traffic with source port 49724 and destination port
56512 to node 2001:db8:1411::24 using CoA1 as source address". The
HA and the correspondent node(s) would "send HTTP traffic with source
port 80 using CoA2 as destination address" and "send voice over IP
traffic with source port 56512 and destination port 49724 using CoA1
as destination address".
A.2.1. Creating new Routing Rules
New routing rules are typically created when an application opens a
new socket. If the mobile node user decides to launch a remote login
session towards CN2 the Connection Manager creates a new routing rule
and decides that this traffic flow should use interface I1. After
this event the rule set at the MN will be as shown below:
tcp peer port 80 on 11
udp local port 49724 peer 2001:db8:1411::24 port 56512 on 800
tcp peer port 22 on 800
The new routing rules must be sent to the HA and optionally to CN2.
Larsson, et al. Expires August 28, 2009 [Page 30]
Internet-Draft Flow Distribution Rule Language February 2009
If for instance the MN user, during the conversation with the CN2
user, gets the possibility to look at some pictures stored at CN2's
computer by using HTTP, then CN2 must be aware of the MN's routing
rules to be able to perform an optimized route selection. The latter
assumes that the MN and CN2 route optimize the traffic sent between
them.
A.2.2. Route Update
If a new interface becomes available or an existing interface becomes
unavailable for some reason, the Connection Manager is notified about
this event and updates the routing rules and the bindings
accordingly. In this example, if the WLAN (interface I2) access gets
out of range the new rule set would look like below:
tcp peer port 80 on 800
udp local port 49724 peer 2001:db8:1411::24 port 56512 on 800
tcp peer port 22 on 800
Since there is only one interface available at the mobile node, a de-
registration of CoA2 and removal of routing rules is needed on the HA
and the correspondent nodes that has the old information stored.
Larsson, et al. Expires August 28, 2009 [Page 31]
Internet-Draft Flow Distribution Rule Language February 2009
Appendix B. Example of Routing Rules
This appendix includes a collection of routing rules giving examples
of how the routing rules can be used to give a feeling for the
language.
Some Generic stuff:
// 6BONE is dead
any peer 3ffe::/16 drop
// Put SSH traffic on low-latency link (uses "any-port")
tcp port 22 on 12
// HTTP is bulk (we don't have web server, so only match peer port)
tcp peer port 80 on 13
// Impress people with short ping times
icmp type 128-129 on 12
// Load share rtp (channel 4 is twice as fast as 5)
udp peer 2001:db8:1c32::/48 port 6900 on 4, 4, 5
// IPSec is asymmetric (SPIs don't match)
ipsec local spi 12 on 18
ipsec peer 2001:db8:8142:500::3775 spi 32 on 18
ipsec on 19
// Some links might be eavesdropped (23 is open, 22 is protected)
esp on 23
tcp port 443 on 23
any on 22
// Bi-cast important traffic
udp peer port 5060 on 11 & 800
// Route on flow label
any local label 4711 on 19
any peer 2001:db8::55 label 99 on 19
// Use traffic class marking
udp tclass 127 on 15
any tclass 128-255 on 14
// Use explicit protocol number
proto 77 on 19
// Rules not matching any address and not having an explicit address
// family will match both IPv4 and IPv6 flows
Larsson, et al. Expires August 28, 2009 [Page 32]
Internet-Draft Flow Distribution Rule Language February 2009
tcp peer port 443 on 11;
udp on 12;
// These two imply the same thing
ip4 tcp port 80 on 12;
tcp peer 0.0.0.0/0 port 80 on 12;
// These are semantic errors (but syntactically OK)
ip4 udp peer 2001:db8::/32 on 10;
tcp local 10.12.14.16/17 peer 3ffe:1234::1 port 99 on 10;
any peer 3ffe:99::1 ttl 12 on 11;
// Drop any previously unmatched IPv4 traffic
ip4 any drop;
// Put all IPv6 traffic on a particular PID
ip6 any on 10;
// Extra
udp local port 99 hop limit 0-22 on 44
any tclass ! 127 ip6h 12-13 drop
Conditional rule sets:
// Depending on the active PIDs different rule sets are used
<11,800>
tcp local port 80 on 800
any on 11
<11>
tcp local port 80 drop
any on 11
Asymmetric routing:
// Send tcp on 19 on uplink, 22 on downlink
tcp on 19 at local
tcp on 22
Mobile networks:
// MNN 2001:db8::77 gets special treatment
udp local 2001:db8::77 port 4400 on 19
udp local port 4400 on 22
Larsson, et al. Expires August 28, 2009 [Page 33]
Internet-Draft Flow Distribution Rule Language February 2009
Authors' Addresses
Conny Larsson
Ericsson Research
Isafjordsgatan 14E
Stockholm SE-164 80
Sweden
Phone: +46 8 404 8458
Email: conny.larsson@ericsson.com
Michael Eriksson
Ericsson Research
Isafjordsgatan 14E
Stockholm SE-164 80
Sweden
Phone: +46 8 757 5888
Email: michael.eriksson@ericsson.com
Koshiro Mitsuya
Keio University
5322 Endo
Fujisawa, Kanagawa 252-8520
Japan
Phone: +81 466 49 1100
Email: mitsuya@sfc.wide.ad.jp
URI: http://www.sfc.wide.ad.jp/~mitsuya/
Kazuyuki Tasaka
KDDI R&D Laboratories Inc.
2-1-15 Ohara
Fujimino, Saitama 356-8502
Japan
Phone: +81 49 278 7574
Email: ka-tasaka@kddilabs.jp
Larsson, et al. Expires August 28, 2009 [Page 34]
Internet-Draft Flow Distribution Rule Language February 2009
Romain Kuntz
Louis Pasteur University / LSIIT
Strasbourg
France
Phone: +33 390 244 584
Email: kuntz@lsiit.u-strasbg.fr
URI: http://clarinet.u-strasbg.fr/~kuntz/
Larsson, et al. Expires August 28, 2009 [Page 35]