Dynamic Memory Allocation in Virtual Machines Based on Cache Hit Ratio

In virtualized environment, several computers run on a physical computer. Many of virtualization systems have a ballooning function with which memory allocation size for a virtual machine can be dynamically changed without restarting the virtual machine. Thus, it is expected that performance of an applications in a virtual machine can be improved by dynamic optimization of the virtual machine memory size. Xen also has a ballooning function, which is called xenballoon. However, it takes account of only memories consumed by processes and does not consider page cache memory size. Thus, I/O performance improvement cannot be expected with this ballooning function. In this paper, we focus on Xen virtualized environment and read-only applications, and discuss a method for improving I/O performance by dynamic optimization of memory allocation size of virtual machines. First, we investigate relation among virtual machine memory size, page cache hit ratio in the guest OS, and I/O performance. Then, we show that providing memory to virtual machines with high cache hit ratio is effective in I/O performance improvement. Second, we propose a method for improving I/O performance based on cache hit ratio. Third, we evaluate our proposed method with filesystem benchmark application FFSB and demonstrate that our method can improve I/O performance by dynamic tuning of virtual machine memory size.