Adaptive Page Migration Policy With Huge Pages in Tiered Memory Systems

To accommodate the growing demand for memory capacity in a cost-effective way, multiple types of are incorporated single system. In such tiered systems consisting small fast and large slow components, accurately identifying performance importance pages is critical to properly migrate hot memory. Meanwhile, address translation cost due increasing footprints, helped adopting huge common systems. Although page hotness identification problems have existed long time, this article revisits problem new context pages. This first investigates locality behaviors applications with three potential migration polices, least-recently-used (LRU), least-frequently-used (LFU), random The evaluation shows that none policies excel others, as effectiveness each policy depends on application behaviors. addition, results show can be effective even migration, if proper used. Based observation, paper proposes novel dynamic selection mechanism, which identifies best given workload. It allows concurrently running workloads adopt different policies. find optimal one workload, study key features must inferred from limited approximate access information collected using accessed bits tables. it parallel emulation alternative assess benefit possible alternatives. proposed achieve 23.8percent improvement compared prior mechanism based LRU lists Linux