Date: Mon, 04 Nov 1996 23:41:47 GMT Server: NCSA/1.5 Content-type: text/html Last-modified: Tue, 18 Jun 1996 12:38:10 GMT Content-length: 3285 Doug Burger's Research Page Research interests:

Research summary:

My research is currently focused on finding ways to improve the performance of uniprocessor on-chip memories. The preliminary part of this research involves exploring the virtues of explicit memory management; memory-mapping part of a cache into the program's address space, or treating it as a large register file. This research is a key component of the Galileo project, which is attempting both to identify fundamental long-term limitations to improved performance and to propose near- and far-term solutions. This is joint work with Alain Kägi, Stefanos Kaxiras, and Jim Goodman.

Related work in which I am currently involved is analyzing the efficiencies of caches for a wide variety of programs. Determining the efficiency of a cache (i.e., how much of the cache contains useful data) and breaking the (in)efficiency into separate categories will hopefully yield insight into some of the fundamental problems with the current caching model.

I am also involved with a beginning effort to evaluate the performance of the Scalable Coherent Interface by directly comparing it with an implementation of the Typhoon protocol processor, through simulation on the Wisconsin Wind Tunnel. Another ongoing project is a study of the scalability of the base SCI protocol (including the standard options), simulating SCI machines comprised of thousands of high-performance processors, running real scientific workloads. This work is being performed on the National Center for Supercomputing Applications' 512-node Thinking Machines CM-5.

Some of my previous work involved studies of SCI, evaluating performance benefits of a range of optimizations (with Alain Kägi) and studying a preliminary design of a MESI/SCI interface chip. I also investigated context scheduling issues for shared-memory multiprocessors, specifically with regard to virtual memory overheads (with Rahmat Hyder, David Wood, and Bart Miller). My main contribution to the Wisconsin Wind Tunnel project was the development of its network simulator and the subsequent evaluation of related issues, performed jointly with David Wood.

Last modified: Tue Jun 18 07:37:00 1996 by Doug Burger

dburger@cs.wisc.edu