Managing cache memory resources in adaptive many-core systems

Abstract

In the last decades, the increasing amount of resources in embedded systems has been leading them to the point where an efficient management of these resources is mandatory, especially for the memory subsystem. Current MPSoCs have more than one application running concurrently. Hence, it is important to identify the memory needs of these applications and provide them accordingly. In this work we propose the use of a cluster-based, resource-aware approach to provide this efficient environment. The solution proposed here improves the overall performance of these systems by aggregating memory resources in clusters and redistributing these resources among applications based on a fairness criterion. For this memory clustering proposal, we use the information of external memory access-es as an estimate of the amount of memory required by each application. T Experimental results show that, depending on how the redistribution of memory resources among application occurs, the overall system can improve performance upto 18% and the energy savings can reach upto 20%.