You can find several different solutions to sprawl in our Virtualization School on how to control VM sprawl. But now it's time for members of our Server Virtualization Advisory Board to offer their favorite strategies as they answer this question:
How much of a problem is VM sprawl in your virtual machine infrastructure (or your customers'), and what are your tips for preventing and eliminating sprawl?
CJ Metz, Orange County United Way
Here in our organization we do not have a large problem with VM sprawl. I believe this is primarily due to the fact that we have a small IT staff and a single individual tasked with VM creation. Thanks to this single point of responsibility, we are better able to manage the systems that are brought online.
Using VMware it's really easy to say, "Hey, we need a new server," and just bring one online without much thought. What we have tried to do, however, is still incorporate a lot of the old methodology used in server creation. For example, even though we could have a bunch of individual application servers, we still utilize global servers, encompassing multiple applications.
Before we purchased VMware Virtual Center Server (VCS), management of our infrastructure was time-intensive and overbearing. I had to log into each host individually and apply whatever changes through that medium. With VMware's VCS I am better able to quickly manage and maintain my virtual servers.
Eric Siebert, Boston Market
VM sprawl is inevitable in medium and large virtual environments unless strict controls are put in place. Just because VMs do not have a physical presence doesn't mean they are free. Every VM has a resource cost.
Implementing measures to deal with VM sprawl should be done early on. Here are some methods for dealing with it:
- Implement a formal process that requires justification for requests for any new VMs. An approval process gets users to think twice about if they really need another VM.
- Utilize resource pools to segregate a host's resources and allocate resources to different groups and departments. This will help control how many VMs are created.
- Implement a chargeback application to provide metrics on host resource usage and to help understand the cost of virtual machines.
- Limit who can create virtual machines, and monitor VM lifecycles so VMs can be deleted when no longer needed.
If you don't control sprawl early on, you may use up all your host resources before you know it and create bottlenecks that may reduce VM performance.
Shannon Snowden, New Age Technologies
The root cause of virtual machine sprawl is the nasty combination of the legacy server management mentality and novice deployment standards.
The baby boom of virtual machines is completely understandable when you compare the ease of VM deployment to the level of effort it takes to provision a physical server. (And when you consider that even that amount of work didn't stop physical server sprawl.)
Here are some tips to avoid virtual server sprawl:
- Have a process where each VM needs approval before deployment.
- Don't over-allocate virtual resources to your VMs. Most physical servers are running at 10% utilization; why repeat the waste in your virtual infrastructure? Your consolidation ratio will suffer.
- Perform post-virtualization performance monitoring on VMs and adjust resources as needed.
- Many third-party utilization tools can pay for themselves by reducing the number of physical assets you require.
Dave Sobel, Evolve Technologies
For us, VM sprawl has not been a problem. We've managed this effectively by using the same criteria we use with physical hardware. There needs to be a business reason for any system deployed, be it physical or virtual, and by enforcing this requirement, you ensure that systems are necessary. Just because you can deploy a VM, that doesn't mean you should!
Once the business need is established, enforcement of a management process around the virtual infrastructure helps ensure that the machines are well managed. Any system, physical or virtual, needs to have a management plan behind it. Ensuring that you track and manage all your deployed systems will keep sprawl under control.
Finally, schedule system review times. Because virtualized systems can be moved, migrated and adjusted much more easily than with physical ones, you can schedule regular reviews of your systems, and schedule any necessary redeployment or consolidation to ensure you don't allow sprawl.
Have a question for the Server Virtualization Advisory Board? Email Colin Steele, Site Editor.
This was first published in December 2009