Internet Engineering Task Force C-Y Lee
INTERNET DRAFT L. Andersson
Expires December 1999 Nortel Networks
Ken Carlberg
SAIC
Bora Akyol
Pluris
June 1999
Engineering Paths for Multicast Traffic using MPLS
<draft-leecy-multicast-te-00.txt>
Status of this memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.
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 view the list Internet-Draft Shadow Directories, see
http://www.ietf.org/shadow.html.
Abstract
This document describes a solution to engineer paths for IP multicast
traffic in a network, by directing the control messages to setup
multicast trees on engineered paths.
This proposal partitions the multicast traffic engineering problem
such that multicast routing protocols do not have to be modified to
setup engineered routes or allocate resources for multicast traffic
nor do resource allocation protocols such as RSVP or CR-LDP have to
be able to setup forwarding states (in this case labels) like
multicast routing protocols.
Resources are allocated on the same trip that paths are selected and
setup. An important aspect of this proposal is that it enables
multicast paths to be engineered in an aggregatable manner, allowing
this solution to scale in the backbone.
Expires December 1999 [Page 1]
Internet Draft Engineering Paths for Multicast Traffic June 1999
1. Overview
In general, traffic is engineered to traverse certain paths so as to
utilize resources in a network in a more optimal manner, while at the
same time improving the level of service that can be offered.
In conventional IP routing, traffic may be engineered to use a path
by configuring preferred links towards a destination with a lower
metric. This method only allows traffic to be engineered based on the
destination address. Since the forwarding is based on the
destination address only, traffic cannot be engineered based on other
attributes (which maybe useful for traffic engineering purposes) of
the packet such as the source address of a packet or the requested
service level. In contrast, MPLS abstracts the forwarding paradigm
and allows traffic to be forwarded based on attributes (known as
forwarding equivalence class (FEC) in MPLS) in addition to the
destination address. This provides a versatile and convenient syntax
for traffic engineering purposes.
This document describes a way to provide a basic traffic engineering
mechanism for multicast. Traffic Engineering (TE) functionalities (in
the MPLS entity) are used to forward the join control messages of
multicast protocols, based on different traffic engineering
requirements and to allocate resources. (Note that multicast data
packets however are forwarded based on Layer 3 (L3) address
information and are not label switched. )
Using this basic multicast traffic engineering mechanism, ISPs can
define particular FECs for their network, resources required to
receive traffic from certain root prefix, decrease fanouts at a node
by limiting the number paths towards the node(prefix), allowing only
certain paths to carry multicast traffic, experiment with heuristics
to better engineer multicast trees, use a function to dynamically
compute suitable paths based on current or predicted network
resources. All these additional network or content provider specific
functions to engineer traffic can be developed independently of the
basic traffic engineering mechanism.
2.0 Motivation
The fundamental problem with doing multicast Traffic Engineering (TE)
is the difficulty in doing it in a scalable manner. Multicast routes
are very difficult (and some claim impossible) to aggregate. One can
associate a label with a unicast route(prefix) and packets sent to
that destination can be aggregated and engineered by associating them
with the label.
Since multicast routes are not aggregatable in general, associating a
Expires December 1999 [Page 2]
Internet Draft Engineering Paths for Multicast Traffic June 1999
label with a multicast route will require per flow/group resource
allocation. In essence, this kind of association will result in RSVP
(or ATM) style resource allocation and is more applicable to per flow
QOS than traffic engineering.
In contrast the approach taken in this proposal decouple traffic
engineering from multicast route setup, thereby allowing the
resources and paths for multicast data delivery to be independently
allocated. What this implies is, resources and paths can be
aggregated and engineered; and traffic can be statistically
multiplexed, enabling network operators to provide differentiated
services for multicast traffic in a scalable manner.
3.0 Scope
This draft described mechanisms which is applicable to multicast
routing protocols such as PIM-SM, CBT, BGMP, Express or Simple
Multicast, which will be called 'control driven' in this draft. 'Data
driven' or flood and prune protocols (eg DVMRP and PIM-DM) are
described in another draft. This proposal assumes a multicast
group/tree has a common 'QOS' requirement. It is envisaged that
heterogeneous receivers requirement can be met by layer encoding data
in different multicast groups or other variation of layer encoding.
It should be noted that the MPLS concepts of interest here are the
FEC, ERO and resource allocation and path selection. An entirely new
supporting protocol could be designed to support the traffic
engineering mechanisms proposed here, however since the concepts of
interest have already been defined and have been implemented in one
form or another, the solution is described in terms of how it can be
realized in MPLS.
4.0 Approach
A control driven multicast routing protocol sends a 'join' message to
graft a node to a multicast distribution tree, creating multicast
routes in the process. Since the join messages are forwarded based on
unicast routes, if the conventional routing table is used, the
multicast routes setup will be based on conventional routes. To
constrain multicast paths, the join message should be sent via paths,
computed or statically configured.
This draft describes a scheme where multicast routing control
messages (including join messages) are forwarded by the MPLS entity
in a router on the constraint path.
To allow a router to process control messages, the control messages
Expires December 1999 [Page 3]
Internet Draft Engineering Paths for Multicast Traffic June 1999
should contain the router alert option. The control message is
identified at the ingress LSR by its FEC. Based on the FEC, the MPLS
entity can derive the path the control message should take and
allocate resources accordingly. A multicast routing protocol would
setup the forwarding state on the ports/interface where the join is
received. To enable the establishment of multicast forwarding state
based on constraint (unicast) routes, multicast routing protocols
which verify the Reverse Path Forwarding (RPF) must turn off this
check. To prevent redundant data and loops, a loop avoidance scheme
based on the concepts described in [MPLS-LOOP-AVOID] or [SM] can be
used in the routing protocol. If there is a loop, the routing
protocol should not create forwarding states for the group on the
port where the join is received.
Other alternatives to send the join on the engineered path such as -
extending CR-LDP/TE-RSVP to send and merge joins for the multicast
tree associated with a label - changing the multicast routing
protocol to send the join along the explicit route, either require
multicast routing protocol functionalities to be present in MPLS or
MPLS functionalities to be incorporated into multicast routing
protocols. This proposal uses MPLS (label and explicit route object)
to cause engineered paths to be selected but forward data using
multicast routing. It does not require MPLS or multicast routing
protocols to be merged, an exercise which tend to - result in
redundant or the reinventing, of functionalities at L2/L3; increase
the complexity of multicast traffic engineering while not providing
any means of aggregating multicast traffic engineering.
The alternative approaches listed above require traffic to be
engineered for each group/tree since multicast labels/routes are most
likely to be not aggregatable. Each group must be assigned a
different label as well. In contrast this proposal allows a network
provider to aggregate the 'QOS' path towards a root or root prefix
(since resource allocaton and path selection can be independent of
the setup of forwarding states/routes). The root prefix could be a
subnet or domain. ulticast traffic in the backbone network can then
be, provisioned in a more scalable manner and statistically
multiplexed on the (aggregated) engineered paths.
5.0 Procedure
5.1 Egress LSR
At any egress LSR (i.e a router where the traffic exits the MPLS
network) that may join multicast trees - FECs, the associated path
selection mechanisms and resources required are specified. These
FECs will match the the control messages of routing protocols (eg
PROTO_ID=PIM-SM/CBT, destination = root prefix/well known multicast
Expires December 1999 [Page 4]
Internet Draft Engineering Paths for Multicast Traffic June 1999
address, TOS=codepoint). Note that the message that carries this
information traverses the network from egress to ingress. The path
selection mechanisms can be based on, a static table or a constraint
based routing table or a path selection algorithm (dynamic). (See
6.0 Path Selection as well)
Figure 1 shows the passage of control messages in an egress LSR
(dotted lines) and the interface between the various entities in the
LSR (+++ lines)
When a control messages arrives at the ingress LSR the packet will be
sent to L3 for processing (where a multicast routing protocol may
setup forwarding states), since the control message contain the IP
Router Alert option. After processing the control message, L3 will
attempt to forward the packet towards the destination specified in
the control message.
------------------------
| Multicast routes |
------------------------
+
+
-------------------------
| Multicast Routing |
-------------------------
^ |
| |
| v
---------- ------------
----> | MPLS | | MPLS | ---->
---------- ------------
+
+
+
---------------
| FEC,Path and |
| Resource |
| Specification |
----------------
Fig. 1 At the egress (wrt data flow) LSR
If the packet (control message) matches the FEC defined in the above
manner, the MPLS entity will invoke the appropriate path selection
Expires December 1999 [Page 5]
Internet Draft Engineering Paths for Multicast Traffic June 1999
mechanism. The root address of the multicast tree may be provided to
the path selection mechanism to obtain the constraint routes towards
the root. The root address of a multicast tree can be retrieved via
a generic API provided by multicast routing protocols. The
constraint routes obtained from the path selection mechanism will be
placed in an ERO. An MPLS control message (CR-LDP/RSVP with MPLS
extension) containing the FEC, ERO TLV, resources required (eg
Traffic Parameter and any other relevant TLVs) will be prepended to
the IP packet. It should then forward the MPLS control message to the
next hop specified in the ERO. To allow routers downstream to
process this control message, the packet will be labeled as Router
Alert.
The explicit routes in the ERO object is removed as it traverses the
explicit path towards the root, in the same manner as described in
CR-LDP and TE-RSVP.
5.2 Intermediate LSRs
Figure 2 shows the passage of control messages in an intermediate LSR
(dotted lines) and the interface between the various entities in the
LSR (+++ lines)
------------------------
| Multicast routes |
------------------------
+
+
-------------------------
| Multicast Routing |
-------------------------
^ |
| |
| v
---------- ------------
----> | MPLS | | MPLS | ---->
---------- ------------
+ +
+ +
+ +
----------------
| FEC |
| State |
----------------
Fig. 2 At an intermediate LSR
Expires December 1999 [Page 6]
Internet Draft Engineering Paths for Multicast Traffic June 1999
When the next hop (or other intermediate nodes) receives the packet
with Label Router Alert, it will be taken out of the forwarding path
and directed to the MPLS entity. (If the control messages are not
labeled, L3 would send this control message directly to a L3
multicast routing protocol, instead of the MPLS entity).
The MPLS entity will allocate the resources requested by the CR-LDP
or RSVP with MPLS extension message, create a state for the FEC (and
other objects eg ERO, Traffic) - called the FEC state for short. It
will then sent the packet to the multicast routing protocol (MRP).
The MRP will then create the forwarding state for the group and will
forward the join message towards the root. Since the FEC for this
control message will match the FEC state created earlier, the join
message will be dispatch to the MPLS entity, which will process the
ERO object and will sent the packet to the next hop listed in the
ERO.
Note that the FEC need only be specified in the ingress LSR,
intermediate LSRs are informed of the FEC information by previous
hops. Similarly, the explicit (constraint) routes is only computed or
configured at the ingress LSR; the next hop and other intermediate
nodes learn of the explicit routes via the ERO object propagated
from the ingress LSR. Loose Source Route can be specified in the ERO
and intermediate nodes (LSRs) may forward it to the next explicit
route/node specified in the ERO based on local routing information.
If an LSR already have an FEC state, the packet will be sent directly
to L3 for processing. L3 will decide if it needs to forward this
control message any further. If it is a join message, and there is
already L3 forwarding states, the join is terminated. If it is a
maintenance control message, the control message is processed and
forwarded. This packet will match the FEC state created earlier and
MPLS will forward the packet according to the next hop in the ERO
list associated with this label and FEC.
5.3 Loops
If the MPLS control message specifies looping explicit routes :
* then if the tree is uni-directional, only the join message will
loop. Data will not loop since data flow is only in one direction
from root to members. * then if the tree is bi-directional, the join
message will loop, but because permanent states would not be
established in this case, data will not be forwarded on the looping
path.
However if there is a change in next hop towards the root at a node
where there is already an existing forwarding state, then multicast
Expires December 1999 [Page 7]
Internet Draft Engineering Paths for Multicast Traffic June 1999
routing protocols which uses bi-directional trees or a hybrid of
uni-directional and bi-directional branches could invoke a loop
avoidance procedure. One way to avoid loops in this case is (using
splice message) described in SM. This procedure should ideally be
specified in the multicast routing protocol itself.
6.0 Path Selection
This proposal allows different path selection algorithms to be used,
depending on the FEC and path selection mechanism association. Paths
can be configured, computed, discovered or obtain through other
means.
A path selection mechanism will return the constraint routes given
for eg the group address, root of multicast tree and other criteria.
How the paths are selected are independent of this proposal, but a
generic interface (API) between path selection algorithms and this
multicast traffic engineering scheme is required and is FFS.
7.0 Examples
This section list some examples of how multicast traffic can be
engineered using the procedures described in this proposal.
a) A network operator may define an explicit route [Rx, Ry, Rz]
towards a domain with prefix 10.0.0.0 for multicast traffic. Any
member joining a group where the root address has the prefix 10.0.0.0
will have data delivered to it via the explicit route [Rz, Ry, Rx]
(data is in the reverse direction of the join control message).
This explicit route may be a Loose Source Route, or a route
calculated by an algorithm eg an Internal Gateway Protocol (IGP)
which can provide constraint based routes.
It is worth noting that the explicit route can be the desired path
from a root towards a member instead of the reverse path (from member
towards the root).
b) Another variation of the above may define an additional field of
interest in the FEC, the TOS. This will allow a network operator, to
allocate resources for traffic belonging to a diffserv forwarding
class, for eg Assured Forwarding.
c) To decrease fanout, egress LSRs (where multicast data traffic
exits) can obtain the contraint routes (via manual configuration or a
constraint based routing entity which can be developed independently
of the basic TE scheme described in this proposal)
Expires December 1999 [Page 8]
Internet Draft Engineering Paths for Multicast Traffic June 1999
d) Load Balancing - a load balancing algorithm can provide the
alternative path that a control message can take depending on the QOS
requirement of the group and the current utilization of the equal
cost paths. As mentioned in the Scope section, this draft assumes the
QOS requirement of the group is constant (or the maximum value is
used) or can be averaged to a constant, for traffic engineering
purposes.
e) Policy routing - Different paths may be defined for different
groups.
8.0 Acknowledgments
The authors are grateful to Dirk Ooms and Yunzhou Li for reviewing
this draft and their helpful suggestions to improve this proposal.
Thanks to Jon Crowcroft for providing insightful comments.
References
[ARCH] E. Rosen, A. Viswanathan, R. Callon, "Multiprotocol Label
Switching Architecture", Work in Progress, July 1998.
[TE-MPLS] Awduche, D. et al., "Requirements for Traffic Engineering over
MPLS", Internet Draft, draft-ietf-mpls-traffic-eng-00.txt, October 1998.
[CRLDP] L. Andersson, A. Fredette, B. Jamoussi, R. Callon, P. Doolan,
N. Feldman, E. Gray, J. Halpern, J. Heinanen T. E. Kilty, A. G.
Malis, M. Girish, K. Sundell, P. Vaananen, T. Worster, L. Wu, R.
Dantu, "Constraint-Based LSP Setup using LDP", Work in Progress,
January, 1999.
[TE-RSVP] D. Awduche, L. Berger, D-H. Gan, T. Li, G. Swallow,
Vijay Srinivasan,
Internet Draft, draft-ietf-mpls-rsvp-lsp-tunnel-02.txt, September 1999
Multicast Routing with resource reservation,
Journal of High Speed Networks 7 (1998) 113-139,
B. Rajagopalan, R. Nair
CBT, Core Based Tree Multicast Routing,
Internet-Draft, March 1998, Ballardie, Cain, Zhang
PIM-SM, Protocol independent multicast-sparse mode Specification,
RFC-2117, June 1997
Estrin, Farinacci, Helmy, Thaler, Deering, Handley,
Expires December 1999 [Page 9]
Internet Draft Engineering Paths for Multicast Traffic June 1999
Jacobson, Liu, Sharma, and Wei.
BGMP, Border Gateway Multicast Protocol Specification,
Internet-Draft, March 1998, Thaler, Estrin, Meyers
Express, H. Holbrook, D. Cheriton
Sigcomm Paper
SM, Simple Multicast, Internet-Draft, March 1999,
draft-perlman-simple-multicast-02.txt, Perlman et al
[MPLS-LOOP-AVOID] "Avoiding Loops in MPLS", Internet Draft,
draft-leecy-mpls-loop-avoid-00.txt, June 1999
C-Y Lee, L. Andersson, Y. Ohba,
YAM, K. Carlberg, J. Crowcroft
Hipparch 1998
Expires December 1999 [Page 10]
Internet Draft Engineering Paths for Multicast Traffic June 1999
Authors' Information
Cheng-Yin Lee
Nortel Networks
PO Box 3511, Station C
Ottawa, ON K1Y 4H7, Canada
leecy@nortel.com
Loa Andersson
Nortel Networks Inc
Kungsgatan 34, PO Box 1788
111 97 Stockholm
Sweden
Phone: +46 8 441 78 34
obile: +46 70 522 78 34
email: loa_andersson@baynetworks.com
Ken Carlberg
SAIC
S 1-2-8
1710 Goodridge Drive
cLean, VA. 22102
Bora Akyol
Pluris Terabit Network Systems
10445 Bandley Drive
Cupertino, CA 95014
USA
akyol@pluris.com
Phone: (408) 861-3302
Fax: (408) 863-0271
email: akyol@pluris.com
Expires December 1999 [Page 11]