Quick memory stream locality analysis

Profiling a memory strewn and a memory system will have an increasing importance in the future as memory systems are getting more complex and the gap between processor and memory speeds increases. This paper proposes simple numerical metrics for measuring both temporal and spatial locality. The developed metrics make a strong simplification of a complex issue, but they provide a convenient method to study and distribute information about locality. To facilitate applying the metrics, a tool for calculating them is presented. Finally, the metrics and the tool are illustrated by examining the locality of the post-cache access stream. The results show how heavier caching decreases the temporal locality of the memory stream, but the spatial locality does not behave as consistently.