Planning virtualization moves wisely
- 26 May, 2008 09:58
Not so long ago, if Underwriters Laboratories needed to add three additional servers worth of computing power in three weeks, the company just bought three new units, says Kent Walker, manager of computer operations for UL. With more time for capacity analysis - which is both labor and time-intensive - Walker might be able to shift resources around and stave off the purchase. But when the need is immediate, there's no time for that.
That's why CiRBA's data center intelligence tool appealed to Walker. The product is designed to automatically analyze a server installation and recommend ways to consolidate hardware using virtual servers instead of physical. Walker asked CiRBA in for a 100-server trial evaluation, to see if the vendor could help rein in his sprawl of underutilized servers.
"They dazzled us," Walker says.
Other tools can tackle pieces of the analysis, such as examining server utilization levels to plan capacity. But CiRBA's tool is comprehensive, letting users factor in technical, business and workload constraints specific to their organizations, analysts say. A financial institution, for instance, might need to stipulate that its trading and research groups not share the same virtual servers due to regulatory requirements.
CiRBA gathers information about the data center in a variety of ways, including agent and agentless discovery mechanisms, and stores the details in a central repository. The tool then applies rules and constraints, builds a multi-dimensional model of the data center and spits out answers on the best way to optimize it, whether it's a physical server farm, a virtual environment, or a mixture of both.
"Very few people have the majority of their servers virtualized, so our customers tend to put it on everything. They get a lot of visibility into both" their physical and virtual environments, says Andrew Hillier, CiRBA's co-founder and chief technology officer.
CiRBA's version 4.6, released last month, added advanced benchmarking to enable users to analyze how specific server workloads would perform on any virtualization platform and fit with other resources. It also built in probability analysis and workload-scoring strategies that factor in the risk and service levels an organization is willing to accept, as well as advanced network and storage analysis.
With those new features, the tool for the first time can be used to model how applications will perform on IBM's System z mainframes and determine which workloads running in Unix- or x86-based environments might be better suited to mainframes.
The upcoming 5.0 version, due in the US summer, will build dynamic models of the data center, tracking daily changes, according to Hillier. A subsequent fall release is expected to provide tighter integration with major system management frameworks, as well as enhanced support for storage analysis, he adds.
Chris Wolf, a senior analyst at Burton Group says other vendors would have "a lot of catching up to do" to match CiRBA's granular capabilities in the planning space. "With other tools, you have to manually account for any non-technical issues, and that can be extremely difficult. CiRBA eliminates a lot of the hard work," Wolf says.
Brad Day, a principle analyst at Forrester Research compares CiRBA's niche in virtual optimization to VMware's commanding position six or seven years ago in virtualization technology. The main alternative to CiRBA's envelope-pushing tool is bringing in consultants to put together Excel spreadsheets, do a presentation and leave, he says.
"The worst thing would be if some big services company bought" CiRBA, Day says.
Customers often opt to participate in CiRBA's 30-day QuickStart Program, where they analyze a subset of their servers to learn how to set up the tool and gain a tangible sense of the potential return on investment they can expect.
When UL enlisted CiRBA for its trial evaluation nearly two years ago, Walker confirmed the ugly picture he feared: far too many servers using less that 10 per cent of their processing power. What he didn't anticipate was how rapidly CiRBA's tool would help with analysis and planning, he says.
Server utilization soared to well over 40 per cent, while the server count dropped by 30 per cent within the first year. Even though Walker still fields the same number of server requests, and the instances of server operating systems grew by 23 per cent, the steady stream of hardware purchases ground to a halt.
"We've gotten there a lot faster than we thought we would, and CiRBA is one of the major reasons," says Walker. "I am shoving servers old servers out the door by the truckload. I don't have any storage space any longer for spare servers."
The next phase of UL's consolidation effort seeks to reduce operating system instances by blending similar workloads. Five Web servers might be condensed to one, Walker says.
"Our confidence is getting higher now that we know how to manipulate the tool to get what we want," Walker says. He notes that one especially helpful feature is CiRBA's ability to assess "what if" scenarios, such as, 'What if I have to reduce count of servers by 10 per cent in three months? What can I do?' CiRBA helps you do it right the first time."