Nick Feamster, MIT
Jennifer Rexford, AT&T Labs
Jay Borkenhagen, AT&T Labs
The Internet consists of about 13,000 autonomous systems (AS's) that
exchange routing information using the Border Gateway Protocol (BGP).
Network operators must have control over the flow of traffic into, out
of, and across their networks. However, BGP does not facilitate
common traffic engineering tasks, such as balancing load across
multiple links to a neighboring AS or directing traffic to a different
neighbor.
Solving these problems is difficult because:
- The number of
possible changes to routing policies is too large to exhaustively test
all possibilities
- Some changes in routing policy can have an
unpredictable effect on the flow of traffic, and
- The BGP decision
process implemented by router vendors limits an operator's control
over path selection.
We use empirical analysis of routing tables and
traffic measurements from the AT&T backbone to derive operational
guidelines for using BGP policy to perform traffic engineering tasks,
such as shifting traffic away from a congested peering link. We show
that a network operator can manipulate traffic efficiently by changing
the routes for the small fraction of destination prefixes (and sets of
related prefixes) responsible for the majority of traffic. We show
how certain BGP policy changes can move traffic in a predictable
fashion, despite limited knowledge about about the routing policies in
neighboring domains.
Finally, we show how operators can gain greater
flexibility for traffic engineering by relaxing some steps in the BGP
decision process (such as the use of AS path length) and ensuring that
neighboring AS's send consistent advertisements at each peering
location. These results enable network operators to use existing
BGP features, such as BGP import policies, to accomplish common
traffic engineering tasks.
An outline of the talk follows:
- Introduction
- BGP Traffic Engineering
- A. Border Gateway Protocol
- B. Tools for network-wide traffic engineering
- Data Collection
- BGP routing tables
- Flow-Level traffic measurements
- Overhead of routing changes
- Simpler import policies: groups of related prefixes
- Fewer routing changes: popular destinations
- Predictability of traffic flows
- More stable traffic volumes: large traffic aggregates
- Inbound traffic: limit globally visible changes
- Tolerating routing changes: limit sensitivity to AS path
- Influence of neighboring domains
- Ensuring consistent advertisements from neighbor AS's
- Limiting the influence of AS path length
- Conclusion
A full draft of our writeup is available from:
http://nms.lcs.mit.edu/~feamster/papers/paper-nanog.ps
[Postscript]
http://nms.lcs.mit.edu/~feamster/papers/paper-nanog.pdf [PDF]
About the Presenter
Nick Feamster is a graduate student in the Networks and Mobile Systems
group at the Laboratory for Computer Science at MIT under the
supervision of Professor Hari Balakrishnan. His research focuses on
security, network video, and wide-area networking. Nick has interned
at HP Labs, Bell Labs, and AT&T Labs. He is an NSF Graduate Research
Fellow and the recipient of the Best Student Paper award at USENIX
Security 2001.
HTML
presentation
RealVideo stream
