Optimized VM memory allocation based on monitored cache hit ratio

Cloud computing by the use of virtual machines has become increasingly important in various situations. In such an environment, multiple virtual machines run on a single physical machine. Many hypervisor implementations have a ballooning function. This function enables virtual machine memory size to be dynamically changed at runtime. Hence, it is expected that dynamic memory resource management while considering the application loads can improve its performance. In particular, I/O performance is expected to be improved significantly because it strongly depends on the size of the HDD cache in an operating system, e.g., a page cache in Linux. The Xen hypervisor has xenballoon, which dynamically changes the size of the virtual machine memory based on the size of memory consumed by the processes in the virtual machines and does not consider the size of the page cache. Therefore, the I/O performance is not improved. In a study, a method for adjusting virtual machine memory size was proposed. However, it assumes applications to be homogeneous. In this paper, we propose a method for dynamic management of virtual machine memory size without assumption for applications. The method takes into account the page cache hit ratio. We show the performance evaluation of the method for various read/write applications. The experimental results demonstrate that our method can improve performance of I/O intensive applications in virtual machines.