Two Optimization Mechanisms to Improve the Isolation Property of Server Consolidation in Virtualized Multi-Core Server Kejiang Ye, Xiaohong Jiang, Deshi Ye, Dawei Huang College of Computer Science, Zhejiang University Hangzhou 310027, China {yekejiang,jiangxh,yedeshi,davidhuang}@zju.edu.cn Abstract—Virtualization brings many benefits such as im- proving system utilization and reducing cost through server consolidation. However, it also introduces isolation problem when running multiple virtual machine workloads in one physical platform. Additionally, with the advent of multi- core technology, more and more cores are built into one die in today’s data center that will share and compete for the resource like cache. It’s worthy to study the isolation of server consolidation in modern multi-core platform. However, to our knowledge there are few work done on the isolation property especially the fault isolation property when one of the virtual machine workloads is attacked in server consolidation. In this paper, we study the isolation property from performance perspective and provide two optimization methods to improve the isolation property. We first define the isolation property and quantify the performance isolation in consolidation and propose a VM-level optimization method. Then we study the fault isolation by introducing a misbehavior virtual machine in server consolidation scenario and propose a core-level cache- aware optimization method to improve the fault isolation. Experimental results show that our two optimization methods can effectively improve the performance isolation and fault isolation with 29.39% and 19.52% respectively. What’s more, Oprofile/Xenoprof toolkits are used to find out the factors affecting isolation property from the hardware events level. Keywords-Cache-aware; Fault Isolation; Multi-core; Server Consolidation; Virtualization I. I NTRODUCTION Virtualization technology [1, 5, 19,20] has brought a lot of benefits such as improving resource utilization, reducing costs, easing management of computer servers and con- solidating multiple server workloads into a single physical platform which named server consolidation. Server consol- idation [3, 4, 13] is one of the most common scenarios of virtualization, which enables to consolidate multiple server workloads into a single physical platform to maximize the system utilization. Although server consolidation offers great potential to improve resource utilization and reduce cost, it may also introduce new challenges in managing the consolidated servers such as isolation. With development of multi-core technology, consolidation of multiple workloads will incur resource competition issues such as core competition and L2 cache competition. What’s more, the scheduling of virtual machines across the physical cores is also a notable issue. In this paper, we will perform a detailed performance study about the isolation property in server consolidation scenario on the multi-core platform, based on which we propose two optimization mechanisms to improve the isolation property. Fig. 1 shows a typical server consolidation scenario in a modern multi-core platform. We divide the isolation property into performance iso- lation and fault isolation. Performance isolation indicates the performance effects when consolidating several work- loads into one physical servers, and fault isolation indicates the performance effects when there exist a misbehavior workload which will affect other workloads. Our goal is to quantify the performance isolation and fault isolation of server consolidation and identify the factors that affect the isolation, and then improve the isolation property using efficient optimization methods based on the our experimental discovery. Fig. 2 shows a consolidation optimization mech- anism to improve the performance isolation. The resource characterization analysis can be done through various meth- ods, such as workloads profiling analysis or tool analysis like Xentop. The main contributions of the paper are summarized as follows: (i) Performance isolation. To explore the isolation under different consolidation scenarios, we formally define the concept of performance isolation and design a variety of experiments using typical workloads in data center to verify its correctness. After the evaluation, one VM-level optimiza- tion mechanism is proposed to improve the performance isolation. (ii) Fault isolation. We also define the concept of fault isolation and try to evaluate it quantitatively by introducing a misbehavior virtual machine with a fork bomb. Based on the experimental discovery, we propose a cache- aware core scheduling mechanism to improve the fault isola- tion. (iii) Profiling analysis. We use the Oprofile/Xenoprof toolkit to profile the hardware events (CPU cycle, L2 cache misses, etc.), helping explain how the multi-core architecture features influence the isolation. The rest of the paper is organized as follows. In Section II, we introduce the background of server consolidation and the motivation to study the isolation property in multi-core platform. Section III describes our evaluation methodology to address the issues of isolation evaluation. In Section IV we 2010 12th IEEE International Conference on High Performance Computing and Communications 978-0-7695-4214-0/10 $26.00 © 2010 IEEE DOI 10.1109/HPCC.2010.95 255 2010 12th IEEE International Conference on High Performance Computing and Communications 978-0-7695-4214-0/10 $26.00 © 2010 IEEE DOI 10.1109/HPCC.2010.95 281