High-Speed TCP Workshop 2007

Purpose

A number of new variants of TCP have been proposed that allow bulk-transfer TCP connections to more efficiently utilize high bandwidth links. However, there is concern that if these variants are deployed on a wide scale, they will adversely impact the performance of the global Internet. There is no consensus on how to evaluate these new TCP variants to ensure that they are safe for widespread deployment. This workshop brings together prominent researchers working in this area to discuss these issues.

We have two specific goals. First, we hope to receive feedback from the participants about Microsoft’s proposed high-speed TCP variant (called CTCP). Second, we will discuss recent results related to testing of these protocols and try to come up with common guidelines about how such protocols should be tested prior to Internet-wide deployment. Ideally, such guidelines will be formalized after further discussion at IETF/IRTF/ICCRG.

Participants

Academia and Industry

Microsoft Windows Networking Division

Microsoft Research

  • Bernard Aboba
  • Deepak Bansal
  • Chris Mitchell
  • Henry Sanders
  • Sandeep Singhal
  • Murari Sridharan
  • Rajesh Sundaram
  • Dave Thaler

Agenda

Monday, February 5

8:45-9:15 Coffee etc.
9:15-9:30 Introductions, agenda bashing, welcome remarks Jitu Padhye
9:30 – 10 Keynote: Enabling High Speed Congestion Control on the Internet Henry Sanders
10 – 10:30 High-Speed Networking: Recent Developments, Issues and Challenges Raj Jain
10:30 – 11 Impact of Background Traffic on Performance of High-speed TCPs Injong Rhee
11 – 11:15 FAST TCP Lachlan Andrew
11:15-11:30 HTCP Doug Leith
11:30 – 12 Compound TCP Murari Sridharan / Kun Tan
12 – 1 Lunch
1 – 1:30 Evaluating new congestion control algorithms Doug Leith
1:30 – 2 Testing TCP over Wide Area Networks Yee-Ting Lee
2 – 2:30 Caltech’s WAN-in-Lab Testbed Lachlan Andrew
2:30 – 3 Break
3 – 3:30 How to standardize new congestion control algorithms Sally Floyd
3:30-5:30 Open Discussion / Panel (details below)
5:30 – 7:00 Break / Head back to hotel / Drive to restaurant
7:00 Dinner

 

Panel Agenda

The panel will focus on discussing and agreeing to a set of criteria for deployment of high-speed congestion control algorithms on the Internet. The guidelines outlined in http://www.ietf.org/internet-drafts/draft-floyd-tsvwg-cc-alt-00.txt could be a starting point for the discussion. The panel can get into questions like a) How is fairness defined? What timescales should fairness be measured? b) is theoretical backing about the stability of the protocol a must? c) Network topologies and traffic mix to test protocols? Is it possible for the research community to agree and publish such a test matrix? d) How are low and high levels of statistical multiplexing on the bottleneck link defined? e) Are some metrics more important than others? f) Should congestion algorithms be tested in isolation from the networking stack? g) do we expect multiple high-speed congestion algorithms to co-exist on the Internet? This might let vendors differentiate themselves as long as the algorithms are safe for deployment.

Tuesday, February 6

9 – 9:30 Coffee etc.
9:30 – 10 An End-to-End Transport Protocol for Extreme Wireless Network Environments K. K. Ramakrishnan
10-10:30 Overview of theoretical research on TCP modeling and stability analysis Peter Key
10:30 – 11 High-speed TCP does not matter to ISPs Albert Greenberg
11 – 11:30 Next steps? Lars Eggert
11:30-12 Concluding remarks Rich Draves
12 – 1 Lunch