Virtual machine performance and capacity management is being bumped up the priority list as IT departments try to squeeze more ROI out of server virtualization,
Many IT departments have taken a reactive approach to virtual machine performance planning -- eyeballing resources for VMs and physical hosts, and adding capacity only when end users complain about performance.
This "gut approach" to VM resource allocation was used in the early days of Kroll Factual Data's server virtualization project, but principle technical architect Christopher Steffen said it became apparent that this "willy-nilly" approach just doesn't cut it when VMs are put in a production environment.
Initially, Steffen's team developed its own resource allocation metrics, based on capacity and fault-tolerance needs, to decide what applications should reside on which VMs and which hosts. This initial effort improved the unit's server utilization rate by 25%, but when it started using Microsoft's System Center Virtual Machine Manager -- which monitors virtual machine performance and physical host resources -- the rate increased by an additional 10% to 15%.
Virtual machine management tools are but one aspect of a bigger picture strategy, however, as Kroll Factual Data, the Loveland, Colo.-based credit report processing arm of risk consultancy Kroll, found out. Steffen's team began virtualizing a few servers, but the project grew -- 650 of its 1,700 physical servers were consolidated down to 22 boxes.
"If you don't think about policies, procedures, a chargeback model and the provisioning model, all those things can come back to bite you in the butt," he said. Those areas don't go away because you virtualized the machine, and as a project gets bigger, you really need to do an impact analysis to really get an idea of how big virtualization can be to the bottom line."
A gut approach to VM resource allocation is still a common practice, said Greg Shields, MVP, VMware vExpert, and founding partner at Concentrated Technology LLC in Denver -- one that will keep you from achieving optimal server consolidation and ROI on your virtualization investment. He instead recommends choosing a VM management tool that combines virtual machine performance and capacity management. "You can't have one without the other -- capacity measures how much resource you have, and performance shows you how well or not well you're using those resources."
Measuring both of these activities is "notoriously difficult" with basic tools that come with server virtualization technologies. "With the basic tools you can get raw performance metrics, but you have to convert that data into something useful, which is not easily done," Shields said.
If you don't think about policies, procedures, a chargeback model and the provisioning model -- all those things can come back to bite you in the butt.
principal technical architectKroll Factual Data
What is needed are more advanced tools that crunch the raw data for you, such as VMware Inc.'s vCenter Capacity IQ tool, VKernel Corp.'s Capacity Analyzer and Microsoft Virtual Machine Manager. Such tools will let you monitor how many VMs you have, and their behavior. "The appropriate tools will preserve SLAs and ensure that the behavior of one VM (using too much resource, for example), does not affect the other VMs," Shields said. "You can create a [resource] reservation for one VM to protect other VMs' service levels."
Ultimately, the tool should help IT balance supply and demand, utilizing real-world metrics. "Midmarket and even smaller firms need some scientific approach if they want to get the most out of their virtualization dollar," he said.
And keep security in mind as you move applications from VM to VM, and host to host, due to changing resource requirements. Certain industries are subject to regulations that do not allow several databases to reside on VMs on the same physical system, or applications to reside on virtual machines that are publicly accessible through a host machine, said Anil Desai, an independent virtualization consultant based in Austin, Texas.
"Monitoring tools do not take those kinds of things into account when they make recommendations," he said. "These tools are great for automating resource allocation, but you just can't automate configurations or resource allocation in a way that allows the tool to move and copy VMs without any oversight."
Virtual machine performance tools will help you make changes based on actual resource usage patterns, to avoid having to fix miscalculation based on a "gut" approach, but such tools must also be reconfigured regularly to realize optimal usage patterns that match the changing needs of the business, he said.
Let us know what you think about the story; email Christina Torode, News Director.