Probability based cache replacement algorithm for the hypervisor cache

Abstract

Virtualization is one of the key technologies which help in server consolidation, disaster recovery, and dynamic load balancing. The ratio of virtual machine to physical machine can be as high as 1:10, and this makes caching a key parameter, which affects the performance of Virtualization. Researchers have proposed the idea of having an exclusive hypervisor cache at the Virtual Machine Monitor (VMM) which could ease congestion and also improve the performance of the caching mechanism. Traditionally the Least Recently Used (LRU) algorithm is the cache replacement policy used in most caches. This algorithm has many drawbacks, such as no scan resistance, and hence does form an ideal candidate to be utilized in the hypervisor cache. To overcome this, this research focuses on development of a new algorithm known as the “Probability Based Cache Replacement Algorithm”. This algorithm does not evict memory addresses based on just the recency of memory traces, but it also considers the access history of all the addresses, making it scan resistant. In this research, a hypervisor cache is simulated using a C program and different workloads are tested in order to validate our proposal. This research shows that there is considerable improvement in performance using the Probability Based Cache Replacement Algorithm in comparison with the Traditional LRU algorithm.