University of Houston
Department of Computer Science

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


Natarajan Ravichandran
will defend his thesis

Making Early Predictions of File Accesses



Abstract
Conventional file access predictors attempt to predict the immediate successor of the file being currently accessed. As a result, they leave almost no time to prefetch the predicted file before it is accessed. We present here a perceptron based file predictor that predicts files to be accessed up to five accesses ahead. It handles changing file access patterns by assigning to each predictor a fixed size operational window during which the predictor for the next window is trained.

Experimental evidence shows that our Early File predictor can make between 50 and 80 percent of correct predictions. While the quality of the predictions decreases when the predictor tries to predict further ahead, this effect remains much less important than that of the system workload.
 

Date: Monday, August 8, 2005
Time: 3:00 PM
Place: 550-PGH


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