Locality-aware cache random replacement policies
Visualitza/Obre
10.1016/j.sysarc.2018.12.007
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/165831
Tipus de documentArticle
Data publicació2019-02
EditorElsevier
Condicions d'accésAccés obert
Tots els drets reservats. Aquesta obra està protegida pels drets de propietat intel·lectual i
industrial corresponents. Sense perjudici de les exempcions legals existents, queda prohibida la seva
reproducció, distribució, comunicació pública o transformació sense l'autorització del titular dels drets
ProjecteCOMPUTACION DE ALTAS PRESTACIONES VII (MINECO-TIN2015-65316-P)
SuPerCom - Sustainable Performance for High-Performance Embedded Computing Systems (EC-H2020-772773)
SuPerCom - Sustainable Performance for High-Performance Embedded Computing Systems (EC-H2020-772773)
Abstract
Measurement-Based Probabilistic Timing Analysis (MBPTA) facilitates the analysis of complex software running on hardware comprising high-performance features. MBPTA also aims at preventing additional analysis costs for timing analysis techniques and preserving the confidence on derived WCET estimates. Cache behavior has a deep influence on WCET estimates and hence on “the amount of software” that can be consolidated onto a single hardware platform. Deterministic replacement policies such as LRU (Least Recently Used) and NMRU (Non-Most Recently Used) have systematic pathological cases that may lead to high execution times and WCET estimates. Instead, random replacement (RR) decreases pathological cases probability, at the cost of temporal locality.
We present two new MBPTA-amenable replacement policies that completely remove the presented pathological cases. The first policy, Random Permutations (RP) preserves higher temporal locality than RR; while the second, NMRU Random Permutations (NMRURP), also protects the Most Recently Used line from eviction. Both proposed policies build upon restricted random replacement choices. Our simulation evaluation (validated against a real prototype) using the Mälardalen benchmarks and a case study shows that RP and NMRURP deliver both high average performance (within 1% of LRUs and NRMU performance) and tight WCET estimates 11% and 24% lower than those of RR.
CitacióBenedicte Illescas, P. [et al.]. Locality-aware cache random replacement policies. "Journal of Systems Architecture", Febrer 2019, vol. 93, p. 48-61.
ISSN1383-7621
Versió de l'editorhttps://www.sciencedirect.com/science/article/pii/S1383762118300912
Col·leccions
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
Locality-aware ... m Replacement Policies.pdf | 942,9Kb | Visualitza/Obre |