A cache topology-aware multi-query scheduler for multicore architectures

Abstract

Growing performance gap between processors and main memory has made it worthwhile to consider off-chip data accesses in multi-query processing [2], [1], [3]. Exploiting data-sharing opportunities among concurrent queries can be critical for effective utilization of the underlying shared memory hierarchy. Given a set of queries, there may be a common retrieval operation for several cases to the same data. A query can benefit from the data previously loaded into the shared cache/memory space by another query. However, if these queries are scheduled independently, it is very likely that the same data is brought from off-chip memory to on-chip caches multiple times, thereby consuming off-chip bandwidth and slowing down overall execution.