A new hybrid approach to exploit localities: LRFU with adaptive prefetching

Authors:
Jike Cui;Mansur. H. Samadzadeh
Affiliations:
Oklahoma State University, Stillwater, OK;Oklahoma State University, Stillwater, OK
Venue:
ACM SIGMETRICS Performance Evaluation Review
Year:
2003

Citing 9
Cited 0

The LRU-K page replacement algorithm for database disk buffering

SIGMOD '93 Proceedings of the 1993 ACM SIGMOD international conference on Management of data
Caching strategies to improve disk system performance

Computer
A data cache with multiple caching strategies tuned to different types of locality

ICS '95 Proceedings of the 9th international conference on Supercomputing
Run-time spatial locality detection and optimization

MICRO 30 Proceedings of the 30th annual ACM/IEEE international symposium on Microarchitecture
On the existence of a spectrum of policies that subsumes the least recently used (LRU) and least frequently used (LFU) policies

SIGMETRICS '99 Proceedings of the 1999 ACM SIGMETRICS international conference on Measurement and modeling of computer systems
Cache Memories

ACM Computing Surveys (CSUR)
Data prefetch mechanisms

ACM Computing Surveys (CSUR)
2Q: A Low Overhead High Performance Buffer Management Replacement Algorithm

VLDB '94 Proceedings of the 20th International Conference on Very Large Data Bases
Performance of LRU Block Replacement Algorithm with Pre-Fetching

MWSCAS '98 Proceedings of the 1998 Midwest Symposium on Systems and Circuits

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper reviewed a number of existing methods to exploit the spatial and temporal locality commonly existing in programs, and provided detailed analysis and testing of adaptive prefetching (a method designed to utilize spatial locality) and the least recently and frequently used (LRFU) method (a method designed to utilize temporal locality). The two methods were combined in this work in terms of their exploitation of locality. The comparative studies of the methods were done using real traces, and hit rate was used as an evaluation measure.Results showed that by using adaptive prefetching, the hit rate improved significantly by an average of 11.7% over the hit rate of LRU in the traces and cache configurations used. It also showed that LRFU consistently gives higher hit rates than LRU, but not by much in the trace files and cache configurations tested. And the X value (a controllable parameter which determines the Weights given to recency and frequency) has to be in a certain range, which is usually narrow, in order to get the best performance for hit rate. Compared to adaptive prefetching and LRU, the hybrid approach of combining adaptive prefetching and LRFU gave a consistently higher hit rate also. But, affected by the performance of LRFU, the improvement in the hit rate by the combination was low.