Hardware/Software Architectures for Real-Time Caching

Abstract

There are two fundamental problems in guaranteeing cache performance for real-time embedded systems: conflict and capacity misses. Though fully associative caches would solve conflict misses, they are too expensive to implement in embedded systems. There are two alternatives: a real-time cache (a software managed fully associative cache with extremely large cache blocks) and a virtually addressed cache. To address capacity misses, one can dynamically (and predictably) manage the cache contents.