A Unified Framework for Hybrid Access Cache Design and Its Applications (1993)
| Venue: | ACAPS Tech. Memo 65, Sch. of Comp. Sci., McGill U |
| Citations: | 1 - 1 self |
BibTeX
@INPROCEEDINGS{Theobald93aunified,
author = {Kevin B. Theobald and Herbert H. J. Humy and Guang R. Gao},
title = {A Unified Framework for Hybrid Access Cache Design and Its Applications},
booktitle = {ACAPS Tech. Memo 65, Sch. of Comp. Sci., McGill U},
year = {1993}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
High-speed microprocessors for high-performance computing demand fast on-chip caches to keep the processor usefully busy. Direct-mapped caches have been utilized for their superior access times, but suffer from higher miss rates due to address conflicts. Recently, this has led to several proposals for hybrid access caches combining the superior speed of direct-mapped caches with the higher hit rates of associative caches. These include the MRU cache [3], the victim cache [6], and the column-associative cache [2]. Contributions of this paper include: ffl A model which unifies these emerging hybrid designs for on-chip caches within a single framework called the Hybrid Access Cache model (HAC model). In this model, a hybrid access cache is conceptually viewed as two caches: a fast direct-mapped primary section, and a slower but associative secondary section, with an efficient mechanism for exchanging data between the two. Under this model, a rich design space is structured in a way whic...
Keyphrases
hybrid access cache design unified framework hybrid access cache direct-mapped cache on-chip cache hybrid design victim cache associative secondary section paper include superior access time mru cache miss rate hybrid access cache model several proposal column-associative cache rich design space superior speed fast direct-mapped primary section associative cache single framework efficient mechanism high-speed microprocessor hac model hit rate
