OSDI '04 Abstract
Pp. 395408 of the Proceedings
Program-Counter-Based Pattern Classification in Buffer Caching
Chris Gniady, Ali R. Butt, and Y. Charlie Hu, Purdue University
Abstract
Program-counter-based (PC-based) prediction techniques have been shown to be
highly effective and are widely used in computer architecture design.
In this paper, we explore the opportunity and viability of applying
PC-based prediction to operating systems design, in particular, to
optimize buffer caching.
We propose a Program-Counter-based Classification (PCC) technique
for use in pattern-based buffer caching that allows the operating
system to correlate the I/O operations with the program
context in which they are issued via the program counters of the
call instructions that trigger the I/O requests. This correlation
allows the operating system to classify I/O access pattern on a
per-PC basis which achieves significantly better accuracy than
previous per-file or per-application classification techniques. PCC
also performs classification more quickly as per-PC pattern just
needs to be learned once. We evaluate PCC via trace-driven
simulations and an implementation in Linux, and compare it to UBM, a
state-of-the-art pattern-based buffer replacement scheme. The
performance improvements are substantial: the hit ratio improves by
as much as 29.3% (with an average of 13.8%), and the execution time
is reduced by as much as 29.0% (with an average of 13.7%).
- View the full text of this paper in HTML and
PDF.
Until December 2005, you will need your USENIX membership identification in order to access the full papers. The Proceedings are published as a collective work, © 2004 by the USENIX Association. All Rights Reserved. Rights to individual papers remain with the author or the author's employer. Permission is granted for the noncommercial reproduction of the complete work for educational or research purposes. USENIX acknowledges all trademarks within this paper.
- If you need the latest Adobe Acrobat Reader, you can download it from Adobe's site.
|