Network Working Group Enke Chen
Internet Draft Redback Networks
Expiration Date: December 2004 Srihari R. Sangli
Procket Networks
Avoid BGP Best Path Transitions from One External to Another
draft-ietf-idr-avoid-transition-00.txt
1. 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.''
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.
2. Abstract
In this document we propose a revision to the BGP route selection
rules that would avoid unnecessary best path transitions between
external paths under certain conditions. The proposed revision would
help the overall network stability, and more importantly, would
eliminate certain BGP route oscillations in which more than one
external paths from one router contribute to the churn.
Chen & Sangli [Page 1]
Internet Draft draft-ietf-idr-avoid-transition-00.txt May 2004
3. Introduction
The last two steps of the BGP route selection (Sect. 9.1.2.2, [1])
involve comparing the BGP identifiers and the peering addresses. The
BGP identifier, (treated either as an IP address, or just an integer
[2]) for a BGP speaker is allocated by the AS to which the speaker
belongs. As a result, for a local BGP speaker, the BGP identifier of
a route received from an external peer is just an random number. When
routes under consideration are from external peers, the result from
the last two steps of the route selection is therefore "random" as
far as the local BGP speaker is concerned.
It is based on this observation that we propose a revision to the BGP
route selection rules that would avoid unnecessary best path
transitions between external paths under certain conditions. The
proposed revision would help the overall network stability, and more
importantly, would eliminate certain BGP route oscillations in which
more than one external paths from one router contribute to the churn.
4. Specification of Requirements
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [5].
5. The Algorithm
Consider the case in which the existing best path A is from an
external peer, and another external path B is then selected as the
new best path by the route selection algorithm described in [1].
When neither Path A nor Path B is eliminated by the route selection
algorithm prior to Step f) - BGP identifier comparison (Sect. 9.1.2.2
[1]), we propose that the existing best path (Path A) be kept as the
best path (thus avoiding switching the best path to Path B).
This algorithm SHOULD NOT be applied when either path is from a BGP
Confederation peer.
In addition, the algorithm SHOULD NOT be applied when both paths are
from peers with identical BGP identifier (i.e., there exist parallel
BGP sessions between two routers). As the peering addresses for the
parallel sessions are typically allocated by one AS (possibly with
route selection considerations), the algorithm (if applied) could
impact the existing routing setup. Furthermore, by not applying the
algorithm, the allocation of peering addresses would remain as a
simple and effective tool in influencing route selection when
Chen & Sangli [Page 2]
Internet Draft draft-ietf-idr-avoid-transition-00.txt May 2004
parallel BGP sessions exist.
6. The Benefits
The proposed revision to the BGP route selection rules avoids
unnecessary best path transitions between external paths under
certain conditions. Clearly the revision would help reduce routing
and forwarding changes in a network, thus help the overall network
stabilities.
More importantly, as shown in the following example, the proposed
revision can be used to eliminate certain BGP route oscillations in
which more than one external paths from one router contribute to the
churn. Note however, that there are permanent BGP route oscillation
scenarios [3] that the mechanism described in this document does not
eliminate.
Consider the example in Fig. 1 where
o R1, R2, R3 and R4 belong to one AS
o R1 is a route reflector with R3 as its client.
o R2 is a route reflector with R4 as its client.
o The IGP metrics are as listed.
o External paths (a), (b) and (c) are as described in Fig. 2.
+----+ 40 +----+
| R1 |--------------| R2 |
+----+ +----+
| |
| |
| 10 | 10
| |
| |
+----+ +----+
| R3 | | R4 |
+----+ +----+
/ \ |
/ \ |
(a) (b) (c)
Figure 1
Path AS MED Identifier
a 1 0 2
b 2 20 1
Chen & Sangli [Page 3]
Internet Draft draft-ietf-idr-avoid-transition-00.txt May 2004
c 2 10 5
Figure 2
Due to the interaction of route reflection [4] and MEDs, the best
path on R1 keeps churning between (a) and (c), and the best path on
R3 keeps churning between (a) and (b).
With the proposed algorithm R3 would not switch the best path from
(a) to (b) even after R1 withdraws (c) toward its clients, and that
is enough to stop the route oscillation.
Although this type of route oscillations can also be eliminated by
other route reflection enhancements being developed, the proposed
algorithm is very simple and can be deployed immediately.
7. Remarks
The proposed algorithm is backward-compatible, and can be deployed on
a per-node basis. The deployment of the algorithm is highly
recommended on a router with multiple external BGP peers (especially
the ones connecting to an inter-exchange point).
Compared to the existing behavior, the proposed algorithm may
introduce some "non-determinism" in the BGP route selection -
although one can argue that the BGP Identifier comparison in the
existing route selection has already introduced some "randomness" as
described in the introduction section. Such "non-determinism" or
"randomness" has not been shown to be detrimental in practice, and
can be completely eliminated by using the existing mechanisms (such
as setting LOCAL_PREF or MED) if so desired.
8. Security Considerations
This extension does not introduce any security issues.
Chen & Sangli [Page 4]
Internet Draft draft-ietf-idr-avoid-transition-00.txt May 2004
9. Acknowledgments
The idea presented was inspired by a route oscillation case observed
on the BBN/Genuity backbone in 1998 while both authors were at Cisco
Systems. The algorithm was also implemented at that time.
The authors would like to thank Yakov Rekhter and Ravi Chandra for
their comments on the initial idea.
10. References
[1] Y. Rekhter, T. Li, and S. Hares, "A Border Gateway Protocol 4
(BGP-4)", draft-ietf-idr-bgp4-23.txt, November 2003.
[2] E. Chen and J. Yuan, "AS-wide Unique BGP Identifier for BGP-4",
<draft-ietf-idr-bgp-identifier-03.txt>, December 2003.
[3] D. McPherson, V, Gill, D. Walton, and A. Retana, "Border Gateway
Protocol (BGP) Persistent Route Oscillation Condition", RFC 3345,
August 2002.
[4] T. Bates, R. Chandra, and E. Chen, "BGP Route Reflection - An
Alternative to Full Mesh IBGP", RFC 2796, April 2000.
[5] Bradner, S., "Key words for use in RFCs to Indicate Requirement
Levels", BCP 14, RFC 2119, March 1997.
11. Author Information
Enke Chen
Redback Networks, Inc.
300 Holger Way
San Jose, CA 95134
e-mail: enke@redback.com
Srihari R. Sangli
Procket Networks, Inc.
1100 Cadillac Court
Milpitas, CA 95035
e-mail: srihari@procket.com
Chen & Sangli [Page 5]
