STI: XCP Development

University Of Southern California, Los Angeles CA

Investigators

Abstract

The purpose of this project is to turn important recent research in TCP congestion control into a practical, validated, and deployable Internet technology. Recent and future changes in transmission technologies and user demands are challenging the fundamental assumptions of the Internet's design, especially the current TCP congestion control. Fiber deployment is soaring, increasing network capacity, and wireless and satellite channels are proliferating, driving up round trip times (RTT). Internet applications, including a variety of scientific and Grid networking applications, are suffering in performance. Need exists to send petabyte files in 15 minutes, i.e., at over 1 Gigabit per second. Technically, large bandwidth-delay product applications perform poorly even in un-congested networks because of TCP congestion control. Large RTT flows get unfair allocations of bandwidth when competing against short flows and TCP has been shown to be unstable when window size becomes very large, causing large bandwidth oscillations. TCP is being patched and tweaked, but a new approach to Internet resource management is needed. The XCP congestion control mechanism that was recently developed by Dina Katabi [Katabi02] fills this need. XCP can exploit rapid changes in available Internet resources in a stable, scalable way, and it can separate link utilization control from bandwidth sharing policies. From these simple capabilities, XCP enables network-friendly data transfer at higher speeds with less jitter and almost no packet loss.The goal of this project is to make XCP real and demonstrate its capabilities and value. This will include full, rigorous design and specification of the protocol, developing a high-speed prototype implementation, testing and validating XCP in a series of increasingly real-life Internet environments, developing migration plans, and moving XCP towards community acceptance and standardization. The proposer will enlist several scientific networking communities to serve as test environments for initial XCP deployment. Final XCP deployment will ultimately take many years. This project will include near-term deployment in research networks and will play a central role in getting this work started - - and in forming a collaborative, community effort for advancing XCP. In particular, the proposer work will both catalyze and leverage the efforts of many groups, including the research community, specific scientific network user communities, router vendors, and the IETF. It is this leverage and synergy, building on their own modest technical effort, that will bring about the generational change in Internet technology that is the central objective of our proposal.