Higher costs also stemmed from the fact that Salsa would have to purchase more OS licenses if it wanted to run VMware, in addition to the cost of obtaining the VMware software itself (VMware did not immediately respond to a request for comment). Because KVM is built into the Linux kernel itself, Salsa could manage virtual infrastructure with its existing RHEL servers. All it needed was a management console, which RHEV provided.
Another benefit with RHEL: Nemmers knew that an upcoming version of KVM would provide the ability to adjust the amount of memory and the number of CPUs each VM would use. This feature could be valuable for Salsa, given that the clients' workloads can vary greatly.
Today, operations of the Washington, D.C.-based Salsa Labs is 70 percent virtualized, running 180 VMs across 50 Hewlett-Packard C-Class blade servers. Only a few very large databases are not virtualized, because they would require additional investment in storage.
"As I end-of-life older hardware, anything running on the middle tier will be virtualized," Nemmers said.
To virtualize the infrastructure, Salsa created a handful of templates based on different functions, such as a Web server template and a database template. Templates can be quickly deployed whenever the functionality is required. In effect, they replace the need for new hardware.
"In situations where we'd normally needed to increase the actual hardware allocation, we'd just virtualize [the function] instead," Nemmers said.
Red Hat got a late start to virtualization game, releasing RHEV in 2009 -- years after VMware and even Microsoft carved out their spots in the market. Whether the company can build its own base of users remains to be seen. Nemmers is not worried about choosing a virtualization package that could very well never be a dominant platform in the field, however.
"KVM has enough of a foothold in the industry. Even if Red Hat de-supported RHEV, I would have no concerns about continuity of operations," Nemmers said. "We could export those virtual guests into our own KVM cluster and be up and running in no time."