University of Houston
Department of Computer Science

In partial fulfillment of the Requirements for the Degree of
Master of Science


Chong Sein Yeo
will defend his thesis

A Group-Based File Predictor


Abstract

Prefetching and caching have been widely used to mask the high latency of disk drives and improve the performance of storage systems. While caching keep in main memory the data items that are the most likely to be reused, prefetching attempts to anticipate user request by fetching data before they are requested.

Most prefetching algorithm attempt to predict the file that is most likely to be accessed next. We investigate the benefits of predicting more than one file in order to eliminate some of the unpredictability of file access patterns. A simulation study performed on seven traces collected at Carnegie Mellon University and University of California, Berkeley has shown the benefits of the approach.

We found that the successor to the currently accessed file X was in 81.4 to 88.2 percent of the cases one of the three last observed successors of file X. Replacing these three most recent successors by the three most frequent successors of Y in 84.8 to 91.3 percent of the cases. As expected, increasing the number of predicted file allows us to correctly predict the successor to the current file in 85 to 90.9 percent of the cases when we select the last five successors of the current file and in 87.6 to 97.9 percent of the cases when we select its five most frequent successors of file Y.

Date: Wednesday, December 13, 2003
Time: 4:30 PM
Place: 550-PGH



Faculty, students, and the general public are invited.
Thesis Advisor: Dr.Jehan0-François Pâris