Title:	Intelligent cache warming
DNr:	SNIC 2019/7-13
Project Type:	SNIC Small Compute
Principal Investigator:	Gustaf Borgström <gustaf.borgstrom@it.uu.se>
Affiliation:	Uppsala universitet
Duration:	2019-02-27 – 2020-03-01
Classification:	10206
Keywords:

Abstract

Doing full system simulations of computer CPUs are often related to capturing and running benchmark programs from certain checkpoints taken during some earlier runtime. As starting a simulation directly from such a checkpoint may most often results in "false" cold cache misses – thus resulting in loss of performance – some initial cache warming is mandatory. Cache warming is a time consuming task in relation to the actual simulation and should thus be minimized. The task of this project is to investigate this problem with the use of the gem5 full system simulator. In addition, our novel Cache Merging technology could be used to speed up simulations even further.