Towards practical page coloring-based multicore cache management

摘要: Modern multi-core processors present new resource management challenges due to the subtle interactions of simultaneously executing processes sharing on-chip resources (particularly L2 cache). Recent research demonstrates that operating system may use page coloring mechanism control cache partitioning, and consequently achieve fair efficient utilization. However, places additional constraints on memory space allocation, which conflict with application needs. Further, adaptive adjustments partitioning policies in a multi-programmed execution environment incur substantial overhead for recoloring (or copying). This paper proposes hot-page approach enforcing only small set frequently accessed hot) pages each process. The cost identifying hot online is reduced by leveraging knowledge spatial locality during table scan access bits. Our results demonstrate identification selective can significantly alleviate coloring-induced adverse effects practice. we also reach somewhat negative conclusion without hardware support, beneficial when performed infrequently (meaning long scheduling time quanta executions).