SEMMA: Secure Efficient Memory Management Approach in Virtual Environment

Abstract

For the ability to explore the memory's fullest potential, memory de-duplication has been widely experimented with in current main stream virtualization platforms. A lot of work has been done to improve the efficiency of memory de-duplication, while too little attention was paid to the introduced security issues which have been proved by prior work. To deal with this security risk and efficiently manage the memory we start our work in this paper. We first conduct extensive experiments on different OS platforms to study the realistic situation of page sharing. By analyzing the results we find: 1) Page sharing is contributed mostly by self-sharing (nearly 90%), and the self-sharing rate varies significantly between Linux and Windows OS platforms. Compared with self-sharing the inter-VM sharing rate is extremely low, under 1% in most cases. 2) Page size has a larger influence on Linux platform than Windows platform, so sub-page de-duplication proposed in prior work may achieve better performance on Linux platform. On the basis of above findings we propose a group-based secure page sharing model (or GSKSM), in which we consider both VM processes and normal processes. We have successfully implemented it in Linux kernel 3.6.6, and the experiment results show it works well with negligible overhead. Finally based on GSKSM we further present an efficient memory management approach (or SEMMA), which combines GSKSM and balloon technique to efficiently manage the memory, and the preliminary experiment result is satisfactory.