ӳ����ý

ӳ����ýies are pros at sharing. They share ideas, data, equipment, and even bikes. So it may be no surprise to learn that behind the scenes of the ӳ����ý’s fast-paced research computing network for data collection and analysis, servers have been quietly getting in the sharing game, too, going “virtual” to save the ӳ����ý money, energy, and space and to keep pace with the growing demand for efficient computing by large and diverse research projects throughout the institute.

When Clint Beilman took on the role of systems administrator three years ago, he was assigned the task of creating a new Windows domain to host many ӳ����ý computers. The job required three computers, but instead of purchasing all new machines (each server costs a few thousand dollars), Clint turned to VMware, a Palo Alto-based company that specializes in virtualization and cloud computation solutions — in other words, “virtual machines” or VMs. By using VM software and re-using some old hardware he found, Clint created a solution with one server. “It was a win-win for the ӳ����ý,” he says.

Servers vastly underutilize their own resources, Clint explains. Unlike the now-debunked myth that humans only use ten percent of their brains, the factoid holds true for servers: the typical machine only uses ten percent of its central processing unit (CPU) on average, and much of the time, usage is closer to one percent. Clint can take advantage of that computational idling by “virtualizing” servers and consolidating them.

A virtual machine is a package of software that acts as though it were a real, physical computer, containing its own software-based CPU, RAM hard disk, and network interface card and running its own operating system and applications. “As far as the operating system knows, it’s a physical machine,” says Clint. On the VMware website, the company makes an even bolder statement: “Even the virtual machine thinks it is a ‘real’ computer.”

Because VMs aren’t tied to one physical machine, they can be consolidated, with many VMs residing on a single, physical server and sharing its resources. The cost savings can be huge, because dozens of VMs can share a single server. For the past several years, Clint has been working to scale up the ӳ����ý’s virtual machine system, which now includes 10 physical servers that host 414 virtual machines, with more added each week.

Clint says that virtual machines seamlessly support diverse work throughout the ӳ����ý’s research platforms and programs, such as genome sequencing, chemical biology, and proteomics, and underlie some of the ӳ����ý’s web-based analytical tools, such as . “They’re used by pretty much everybody here,” he says, “even though not everybody knows they’re using it.”

The advantages of virtual machining are many. In addition to being economical, says Clint, “they’re green.” By only taking up one rack in the server room, the VMs cut down on cooling and power needs, a major expense in the computing arena. “That was initially why I started doing this,” he says.

Another benefit of virtual machines is known as high availability. Because the VM hosts have so much capacity (each VM host server has 256 gigabytes of memory and 24 cores, or CPUs), they can provide refuge for VMs whose host server malfunctions or crashes. The fluidity of the system also comes in handy when hardware work is needed. Clint can put a host into maintenance mode, which flushes all its VMs to another host, enabling him to work on the hardware without disturbing the work of the virtual machines. Maintenance and repair of traditional hardware is often done at night or on weekends to minimize the disruption to research. “I can now change out a whole cluster during the week, during normal working hours, without any disruption to scientific research,” Clint says.

Virtual machines are so successful that Clint and his team are now scaling up the system. “Most researchers prefer virtual machines to physical ones, due to their lower acquisition cost and higher availability of services,” says David Parrella, a team leader in the research computing group. Clint's biggest challenge this year will be to keep up with demand from the ӳ����ý's scientific community as the scale of sequencing and genome research – and associated computing needs – continue to grow. David adds, “It’s a challenge for which Clint is uniquely qualified to meet.”