Application Performance Isolation in Virtualization Gaurav Somani and Sanjay Chaudhary Dhirubhai Ambani Institute of Information and Communication Technology, Gandhinagar, INDIA {gaurav_somani, sanjay_chaudhary}@daiict.ac.in Abstract Modern data centers use virtual machine based implementation for numerous advantages like resource isolation, hardware utilization, security and easy management. Applications are generally hosted on different virtual machines on a same physical machine. Virtual machine monitor like Xen is a popular tool to manage virtual machines by scheduling them to use resources such as CPU, memory and network. Performance isolation is the desirable thing in virtual machine based infrastructure to meet Service Level Objectives. Many experiments in this area measure the performance of applications while running the applications in different domains, which gives an insight into the problem of isolation. In this paper we run different kind of benchmarks simultaneously in Xen environment to evaluate the isolation strategy provided by Xen. Results are presented and discussed for different combinations and a case of I/O intensive applications with low response latency has been presented. 1. Introduction Virtual machines are the key blocks of utility computing or on-demand facilities like cloud computing. Modern data centers host different applications ranging from web servers, database servers and high performance computing nodes to simple user desktops. Although the concept of virtualizing resources is three decades old [1] but it is gaining popularity after the term on-demand computing or cloud computing arose. There are types of virtualization technologies available: Full virtualization, Para-virtualization and OS level virtualization These techniques differ from each other in their internal architecture and how they communicate with guest operating systems. Xen is the open source virtual machine monitor developed at computer laboratory, University of Cambridge, UK. It follows para-virtualization methodology in resource virtualization [2]. Data centers which host these virtual machines on their physical machines follow Service Level Agreements (SLAs), which specifies the service requirements with different constraints and parameters to be fulfilled by service provider or cloud provider [3]. These constraints and parameters include total uptime and downtime, requirement of CPUs, network bandwidth and disk space. While running more than one virtual machine on a single physical server, virtual machine scheduler is responsible for allocating resources as defined by SLAs. This allocation also includes a most demanding and inherent property which is referred as isolation among virtual machines. Isolation is meant for securing and providing the resources to a virtual machine which is co-hosted with other virtual machines on a single physical server. These resources are CPU share, network share, memory share and disk share to each virtual machine. Thus isolation property is forbidding a misbehaving virtual machine to consume other virtual machine resources and providing fairness according to their shares. In this paper we are intended towards checking isolation using a set of experiments on Xen virtual machine monitor. These experiments are aimed towards getting a perception of scheduling granularity and their effects on applications. Section 2 of this paper discusses the Xen architecture and scheduling algorithms provided. Section 3 elaborates experimental setup and their relevance. Section 4 discusses the results and their significance in the isolation problem. Section 5 discusses related work in research community and Section 6 concludes and directed towards future work. 2. Xen Virtual machine monitor 2.1 Architecture Xen is the virtualization tool for the x86 architecture which supports paravirtualization. To support full virtualization it requires virtualization technology enabled processors. Xen architecture shown in figure 1 elaborates the basic blocks in it. Xen designates domain-0 which is the host operating system as isolated driver domain (IDD) to provide device driver support to guest operating systems. Thus in Xen architecture the device drivers in host 2009 IEEE International Conference on Cloud Computing 978-0-7695-3840-2/09 $25.00 © 2009 IEEE DOI 10.1109/CLOUD.2009.78 33 2009 IEEE International Conference on Cloud Computing 978-0-7695-3840-2/09 $25.00 © 2009 IEEE DOI 10.1109/CLOUD.2009.78 33 2009 IEEE International Conference on Cloud Computing 978-0-7695-3840-2/09 $26.00 © 2009 IEEE DOI 10.1109/CLOUD.2009.78 33 2009 IEEE International Conference on Cloud Computing 978-0-7695-3840-2/09 $26.00 © 2009 IEEE DOI 10.1109/CLOUD.2009.78 41 2009 IEEE International Conference on Cloud Computing 978-0-7695-3840-2/09 $26.00 © 2009 IEEE DOI 10.1109/CLOUD.2009.78 41