Measuring Up for Server Virtualization

Server virtualization is one of the hottest topics for the capacity planner. How does a hardware virtualization environment like VMware ESX impact the performance measurements we've been accustomed to getting from Windows and Linux? If you can't measure it, you can't manage it, so how are we to measure and understand this new environment? This paper presents a lab-based study designed specifically to aid in understanding performance measurements available directly from VMware ESX. This study sheds light on the impact of virtualization on Windows performance counters. INTRODUCTION Server virtualization is one of the hottest topics in the IT industry. Gartner Inc. Vice President John Erick describes virtualization as a " megatrend." Megatrend may sound like a pretty strong word, but from a capacity planner's perspective, we should consider it good news for our careers. Virtualization is hot because there are very real and quantifiable dollar savings to be gained through proper server consolidation. That's not a misprint. Server virtualization is a technological method to consolidate servers. For those of you with a mainframe background, consider the business benefits of server consolidation mirroring the advantages of logically partitioning the mainframe. Server consolidation eliminates out-of-date hardware, reduces systems administration overhead, simplifies disaster recovery scenarios and can improve overall availability. There are other benefits—nearly all of which translate into real dollar savings. If you don't believe it, do some web searches and you'll find plenty of solid case studies that prove it. Here is the good news: To be successful with server virtualization requires some solid capacity planning discipline. It has been quite a while since an IT industry " megatrend " was good for the capacity planners, but we've got ourselves a winner in server virtualization. The purpose of this paper is to introduce some of the basic issues involved with measuring and reporting performance information from a virtual server environment. The issues will be explored by running a series of lab-based experiments and extracting measurements from various observation points.