Transaction-aware SSD Cache Allocation for the Virtualization Environment

Abstract

Flash-based Solid State Disk (SSD) is widely used in the Internet-based virtual computing environment, usually as cache of the hard disk drive-based virtual machine (VM) storage. Existing SSD caching schemes mainly treat the VMs as independent units and focus on critical performance metrics concerning one single VM, such as the IO latency, throughput, or the cache miss rate. However, in the Internet-based virtual computing environment, one transactional application usually consists of multiple VMs on different hypervisors. Transaction-aware SSD caching schemes may potentially better improve the end user-perceived quality of service. The key insight here is to utilize the relationships among VMs inside the transactional application to better guide the allocation of the SSD cache, so as to help learn the pattern of workload changes and build adaptive SSD caching schemes. To this end, we propose the Transaction-Aware SSD caching (TA-SSD), which takes the characteristics of transactions into consideration, uses closed loop adaptation to react to changing workload, and introduces the genetic algorithm to enable nearly optimal planning. The evaluation shows that comparing to the equally partitioned cache, the allocation produced by the TA-SSD can boost the performance by up to 40%, with dynamic changes in the intensity and the type of the workload.