CASH: A Context-Aware Scheduling for Hypervisor Based on KSM

Cache locality, or cache hit rate, is closely related with the system performance. Meanwhile the trade off between performance and fairness is also needed to be considered by OS at the system level. Hence it is a complex problem for hyper visor scheduler to make use of cache locality effectively. In fact, cloud does bring new opportunities to this problem due to its new application characteristics. In this paper, we propose a context-aware scheduling for hyper visor, based on Kernel Same page Merging (KSM) and the assistance of modern hardware, to improve system performance, making use of context similarity among VMs with similar function. To the best of our knowledge, CASH is the first scheduler to find context similarity based on KSM, and has little to no impact on the scheduling flexibility.