VM sprawl, or virtual machine sprawl, is just what it sounds like: too many VMs sprawled across a virtual infrastructure, taking up processing power and storage space even if they are rarely used. Since spinning up a new VM can be done in a matter of minutes, users come to expect a new machine, on-demand, whenever they want it.
And that's not necessarily a bad thing. IT exists to provide services to business users, and user requests are, more often than not, reasonable.
"Everybody makes reasonable requests," Turner says. The challenge for IT is, "how do you take what are reasonable request for resources and try to align those into the bigger picture?"
The problem that presents is not just related to usage of storage and network resources; each VM adds to the time limited IT staff must devote to managing virtual infrastructure, Turner notes.
Sabre Holdings, a VMware customer in the online travel business located in Southlake, Texas, has about 2,000 VMs in its test and development lab and another 100 VMs in production. It's definitely a case of sprawl in the development lab, says Glenn Harper, Sabre's chief infrastructure architect. The focus for Harper and his IT colleagues has mainly been providing new functionality to developers, but now they are planning to implement chargeback and capacity products to identify VMs that can be turned off.
With a small virtualization deployment, a few conversations with developers might be enough to solve the problem, but not when you reach the level of 2,000 VMs.
"There's just no way a single person can sort through that and get their day job done," Harper says.
Turner says he isn't a fan of chargeback and metering systems "because they create artificial barriers," but he does use VMware management tools to monitor VM usage and he says it's important to educate users about the cost of running VMs. In some cases, if a user wants a new VM, Turner may propose a "swap" in which an older VM, one that isn't heavily used, is turned off.
"We try to educate them that there is a real cost," he says. "What we try to say is it's really a human resources problem.
Managing 800 machines is not like managing 400 machines. It's not double the work, it's way more than double the work. At some point, you run out of human resources to deal with it and your risk starts to increase."