Virtualization is a buzzword right now, and it’s no wonder agencies are coming around to the idea of consolidating their servers and storage. Traditional servers do nothing for about 80 percent of their lifecycle, yet they use nearly half their peak energy consumption — a waste of capacity and power.
Server virtualization creates logical machines on a single physical server. At the Energy Department’s Pacific Northwest National Laboratory in Richland, Wash., virtualization technology is proving to be a cost-effective way to make better use of server hardware resources while reducing hardware lifecycle costs and cooling demands, and saving precious data-center space. Virtualization also ties in with the lab’s commitment to be a responsible steward of the environment and DOE assets.
Based on that experience, here are insights on the PNNL experience and tips on how even the smallest IT shops can successfully go virtual.
Taking the Leap
The need to maximize storage capacity and minimize storage management costs led the lab’s Information Technology Services Division to deploy storage virtualization of its business management systems in 2001. With more than 60 terabytes virtualized today, the technology has become one that PNNL uses daily and would be reluctant to do without.
The lab, part of the DOE Office of Science, works on fundamental science and technology development in the areas of energy, environment and national security. It runs advanced systems for scientific computation and relies on a vast array of computational resources to conduct programs that bring together researchers from government, academia and industry. To support that work, IT Services manages multiple business applications and back-office systems.
The IT team found that storage virtualization contributed to a significant reduction of downtime experienced by the lab’s users during storage reconfiguration and relocation of these systems. System outages, even those on the weekend, can significantly affect staff productivity and the perceived reliability of IT service organizations.
For instance, PNNL needed to relocate a storage array supporting 16 terabytes of storage and 21 servers to another data center. By using storage virtualization features, the lab reduced required downtime to a single reboot per server and total outage for the move to five hours, compared with an estimated 126 hours for traditional dedicated storage. That same event today could occur with no downtime at all. Administrators can now accomplish tasks during business hours that traditionally were done on nights and weekends.
Based on the success of storage virtualization, the IT Services Division began eyeing similar virtualization technologies with the hope of eliminating a large number of physical servers.
After evaluating virtualization technologies and building an environment to support virtual servers, the team rolled out a server virtualization pilot project in late 2005. As summer 2006 approached with record-breaking temperatures, the need to accelerate the pilot literally heated up when power, cooling and floor space reached critical limits in the primary data center while the IT Services Division awaited planned capacity upgrades.
Electrical breakers tripped, servers went down, and suddenly, the crunch on space and resources forced the team to go full speed ahead with the virtualization effort. This decision paid off immediately. By virtualizing 69 management system servers into 12 physical servers — a six-to-one reduction — power and cooling demand fell 7 percent. The team saved PNNL more than $80,000 in 2006; savings in 2007 doubled to nearly $175,000.
Moreover, virtual server relocation and system maintenance can be accomplished with no impact on the users. Less physical server maintenance means increased availability for them, while memory replacement or other hardware issues have become transparent. Through live migration, PNNL frequently performs hardware maintenance and patching on the virtualization layer of the physical hosts without causing any application or IT service downtime.
Another Set of Eyes
Choosing which servers to virtualize first takes careful consideration. PNNL has made it a best practice to choose good candidates based on resource requirements. Often, a server that is quietly doing its job is a better candidate than a more demanding, resource-intensive system, such as one running a large relational database.
A CPU-intensive server can take twice the time and expense to virtualize than it would take to virtualize four or five other servers. In that case, the right decision might be to leave the demand-intensive application running on a physical server. As virtualization technologies mature, even resource-intensive, relational databases become possible candidates. But waiting to virtualize heavy workloads until last remains a prudent course of action so that your team can gain experience. Currently, any new server or server needing life-cycle replacement is a candidate for virtualization.
It is valuable to have another team review the proposed policies and practices when undergoing an ambitious virtualization project. Halfway through the first project, PNNL contracted an experienced third-party company to assess its physical servers. The assessment not only validated the lab’s choices but also pointed out additional physical servers that would be excellent candidates for virtualization.
Green and Lean
Today, the lab’s IT Services Division runs 136 virtual machines on 23 physical servers and plans to virtualize 40 more in 2008. The group manages 400 servers for all types of management systems, including payroll, finance and human resources. (Scientific computing virtualization at the lab is conducted under a separate effort.)
But virtualization isn’t just for large and midsize organizations; small IT shops can save money and resources by virtualizing, too. Consider this: Instead of investing in eight physical servers, your agency could purchase two and run virtual servers on them.
You will get the best benefit if you have a storage area network. Without a SAN you might sacrifice features such as live migration between hosts, but you will still get the benefit of using fewer physical servers to host your server environment.
The PNNL team has measured the greatest bang for the buck in cost avoidance, lifecycle expenses and savings for the demand on power and cooling. With so many paybacks, every IT shop should consider some level of server — and storage — virtualization.