HHMO: A Hot-page Based Hybrid-copy Migration Optimization Method

摘要

The post-copy and hybrid-copy migration can not cope with remote page fault, leading to decline of quantity of service (QoS) especially in memory intensive workload. HHMO: a Hot-page based Hybrid-copy Migration Optimization is a highly optimized pre-paging method proposed to solve the remote page fault problem. This paper firstly defines "hot-page" in migration as the re-modified and re-accessed memory page, and "hot-area" as a contiguous address space covering a lot of hot-pages, then shows the process of HHMO: it captures the hot-areas, recognizes the workload type at the same time, then saves hot-area information to improve prediction accuracy of memory access pattern, finally pushes these page contents into transmission queue to finish pre-paging. Further, a novel evaluation metric: suspending time is proposed for quantification of performance degradation caused by remote page fault in both post-copy and hybrid-copy migration. Eventually, the results of simulation and benchmark experiments demonstrate the effective of HHMO method.