skip to main content
Department of Computer Science University of Colorado Boulder
cu: home | engineering | mycuinfo | about | cu a-z | search cu | contact cu cs: about | calendar | directory | catalog | schedules | mobile | contact cs
home · events · thesis defenses · 1998-1999 · 
 

Thesis Defense - Maltzahn

 
4/16/1999
3:30pm-5:30pm
ECOT 831

Improving Resource Utilization of Enterprise-Level World-Wide Web Proxy Servers
Computer Science PhD Candidate

The resource utilization of enterprise-level web proxy servers is primarily dependent on network and disk I/O latencies and is highly variable due to a diurnal workload pattern with very predictable peak and off-peak periods. Often, the cost of resources depends on the purchased resource capacity instead of the actual utilization. This motivates the use of off-peak periods to perform speculative work in the hope that this work will later reduce resource utilization during peak periods. We take two approaches to improve resource utilization.

In the first approach we reduce disk I/O by cache compaction during off-peak periods and by carefully designing the way a cache architecture utilizes operating system services such as the file system buffer cache and the virtual memory system. Evaluating our designs with workload generators on standard file systems we achieve disk I/O savings of over 70% compared to existing web proxy server architectures.

In the second approach we reduce peak bandwidth levels by prefetching bandwidth during off-peak periods. Our analysis reveals that 40% of the cacheable miss bandwidth is prefetchable. We found that 99% of this prefetchable bandwidth is based on objects that the web proxy server under study has not accessed before. However, these objects originate from servers which the web proxy server under study has accessed before. Using machine learning techniques we are able to automatically generate prefetch strategies of high accuracy and medium coverage. A test of these prefetch strategies on real workloads achieves a peak-level reduction of up to 12%.

Committee: Dirk Grunwald, Associate Professor (Chair)
Dennis Heimbigner, Research Associate Professor
James Martin, Associate Professor
Gary Nutt, Professor
Kathy Richardson, Digital Equipment Corporation

 
See also:
Department of Computer Science
College of Engineering and Applied Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
Questions/Comments?
Send email to

Engineering Center Office Tower
ECOT 717
+1-303-492-7514
FAX +1-303-492-2844
XHTML 1.0/CSS2 ©2012 Regents of the University of Colorado
Privacy · Legal · Trademarks
May 5, 2012 (13:40)
 
.