Title :
Utility-based dynamic cache resizing
Author :
Tian, Geng ; Liebelt, Michael
Author_Institution :
Sch. of Electr. & Electron. Eng., Univ. of Adelaide, Adelaide, SA, Australia
Abstract :
Continuing increases in transistor density are enabling increases in the number of on-chip cores and the size of caches. However very large caches in future systems might not be fully utilised. Under-utilised active caches result in a waste of static power. Off-line profiling of applications to determine optimal cache size and configuration is not practical. As an alternative we propose a simple scheme with limited hardware overhead, to dynamically evaluate the utility of the last level cache slice on each tile and to tune the cache associativity at tile-level granularity. In simulations our tuning scheme achieved an average of 30% of static power saving with a slight 0.6% degradation of IPC.
Keywords :
cache storage; power aware computing; IPC; cache associativity; cache slice; hardware overhead; on-chip cores; static power saving; static power waste; tile-level granularity; transistor density; tuning scheme; under-utilised active caches; utility-based dynamic cache resizing; Monitoring; Multicore processing; Radiation detectors; Resource management; Tiles; Transistors; Tuning; Cache; Multi-threading; Power gating; Static power saving; Tiled structure;
Conference_Titel :
Computer Science and Network Technology (ICCSNT), 2011 International Conference on
Conference_Location :
Harbin
Print_ISBN :
978-1-4577-1586-0
DOI :
10.1109/ICCSNT.2011.6182032