The unconventional replacement algorithms

Authors:
Humayun Khalid
Affiliations:
Department of Electrical Engineering, Convent Avenue at 140th street, City University of New York, The City College, New York, NY
Venue:
ACM SIGARCH Computer Architecture News
Year:
1995

Citing 4
Cited 1

Introduction to the theory of neural computation

Introduction to the theory of neural computation
A methodology for evaluating the performance of CISC computer systems under single and two-level cache environments

Microprocessing and Microprogramming
Cache Memories

ACM Computing Surveys (CSUR)
Analysis of cache replacement-algorithms

Analysis of cache replacement-algorithms

Experimental studies of access graph based heuristics: beating the LRU standard?

SODA '97 Proceedings of the eighth annual ACM-SIAM symposium on Discrete algorithms

Quantified Score

Hi-index	0.00

Visualization

Abstract

This work proposes several new schemes for the replacement of cache lines/blocks in high performance computer systems. Our algorithms rely on judiciously chosen neural networks for accurate real-time statistical predictions. These algorithms, therefore, provide better cache performance as compared to the conventional LRU (Least Recently Used) algorithm. Simulation results indicate that the proposed set of replacement strategies can provide a performance improvement of as much as 16.4711% over the LRU algorithm for our selected benchmark trace files. The results are based on an experimentation involving 6 neural network paradigms and 21 different cache configurations. Excellent performance of the neural network-based replacement strategies means that this new approach can be studied as an alternative to the existing page replacement and prefetching algorithms in virtual memory systems.