MBSA: a Lightweight and Flexible Storage Architecture for Virtual Machines
Xian Chen,Wenzhi Chen,Zhongyong Lu,Yu Zhang,Rui Chang,Mohammad Mehedi Hassan,Abdulhameed Alelaiwi,Yang Xiang
DOI: https://doi.org/10.1002/cpe.4028
2017-01-01
Abstract:SummaryWith the advantages of extremely high access speed, low energy consumption, nonvolatility, and byte addressability, nonvolatile memory (NVM) device has already been setting off a revolution in storage field. Conventional storage architecture needs to be optimized or even redesigned from scratch to fully explore the performance potential of NVM device. However, most previous NVM‐related works only explore its low access latency and low energy consumption. Few works have been done to explore the appropriate way to use NVM device for improving virtual machine's storage performance. In this paper, we comprehensively evaluate and analyze conventional virtual machine's storage architecture. We find that, even with cutting‐edge optimization technologies, virtual machine can only achieve 30% of NVM device's original performance. Based on this observation, we propose a memory bus–based storage architecture, which we named MBSA. Memory bus–based storage architecture can greatly shorten the length of virtual machine's storage input/output stack and improve NVM device's use flexibility. In addition, an efficient wear‐leveling algorithm is proposed to prolong NVM device's lifespan. To evaluate the new architecture, we implement it as well as the wear‐leveling algorithm on real hardware and software platform. Experimental results show that MBSA can provide a big performance improvement, about 2.55X, and the wear‐leveling algorithm can efficiently balance write operations on NVM device with a negligible performance overhead (no more than 3%).