Network performance in tightly-coupled multiprocessors typically
degrades rapidly beyond network saturation. Consequently, designers must keep a network below its
saturation point by reducing the load on the network.
This becomes even more important with advent of multiple hardware
contexts on a chip (e.g., single-chip multiprocessors or simultaneous
multithreading) that can dramatically increase network load.
Congestion control via source throttling---a common technique to reduce
the network load---prevents new packets from entering the network in the
presence of congestion. Unfortunately, prior schemes to implement source throttling
either lack vital global information about the network to take the correct
decision (whether to throttle or not) or depend on specific network parameters,
network topology, or communication patterns.
The goal of this research is to develop self-tuned, global-knowledge
based congestion control techniques for multiprocessor networks.
People
Alvin R. Lebeck (Professor Duke Computer Science)
Shubhendu Mukherjee (Compaq VSSAD)
Mithuna Thottethodi (PhD Candidate Duke Computer Science)
Publications
Self-Tuned Congestion Control for Multiprocessor Networks, M. S. Thottethodi, A. R. Lebeck, S. Mukherjee, to appear Seventh International Symposium on High Performance Computer Architecture (HPCA-7), January 2001 (ps)
Funding
NSF, Compaq, Intel