Having implemented virtualization software to conquer "server sprawl" (or, more specifically, to perform more work on fewer servers and thus minimize the proliferation of servers), many Wall Street firms are now in the throes of "virtual sprawl." This panicky term refers to the condition in which IT managers and, in some cases, even end users have been installing virtual machines all over the organization, creating a chaotic mess that is hard to manage. As a result, IT resources aren't being allocated appropriately, meaning important applications may not be getting the server, storage and memory resources they need. Some virtual machines may even get lost or forgotten altogether.
"Everybody is moving aggressively in virtualization," notes Julien Courbe, managing director, financial services technology, at BearingPoint. "Wall Street is no exception."
Virtual sprawl "is a problem most companies create for themselves," adds Nels Marin, managing director of New York-based Financial Services Consulting. "Virtualization is a very efficient way to use hardware, but it's hard to keep track of when you have thousands of servers and technicians and users who, in some cases, access these servers more directly than you'd like them to."
The old way of doing things -- hardware server deployments -- had some controls naturally built in that have been lost, according to Jacob Hall, chief architect for the investment bank at Wachovia. "As we virtualize everything, virtual sprawl becomes an issue and concern for us," he says.
The Downside to Virtualization
For one thing, virtual sprawl can affect software licensing costs, Hall notes. "If you don't put a process around ordering equipment, rightsizing it and ensuring that operating systems are installed in the correct place relative to other systems they talk to, and if you let users deploy virtual machines without proper chargeback allocation models, you may end up buying more software than you really need," he explains. Further, software vendors increasingly are policing their customers, auditing how many instances of their programs a company is really running versus the number of licenses it purchased and then asking the client to cough up the difference, which can run into millions of dollars.
In addition, virtual sprawl can affect IT's ability to fix problems. "Because you have more than one operating system on a single physical machine, and because virtual machines can move around dynamically, it becomes even more complex to troubleshoot," Hall says. "Virtualization requires more diligent monitoring and management."
Another potential problem is over-virtualizing and consequently failing to give applications the processor, storage and memory resources they need. "There's a temptation to over-subscribe the hardware," Hall relates. "People need to be practical about how they use virtualization. Just because you can stack a thousand operating systems on a single machine doesn't necessarily mean it's a great idea -- if that machine fails, you get a thousand phone calls, and a thousand trouble tickets are generated."
Even a conservative virtualization ratio, such as two to four virtual machines to every physical machine, can provide significant benefits, Hall points out. "And because you have fewer virtual machines running on the same physical machine, those contention issues are not as much of an issue."
Critical programs -- such as algorithmic trading, financial modeling and simulation software -- can suffer from virtual sprawl, according to Jonathan Cohn, a consultant in the strategic IT and operations practice at Oliver Wyman. "At some clients, we've seen a lack of communication and management between applications, such as a large-scale derivatives platform running Monte Carlo pricing simulations," and the underlying virtual machines, he says. "It's not trivial to map the processing needs to an application and then tune it. It's not as simple as vendors would have one believe."
So why not let important, time-sensitive applications run on dedicated hardware instead of putting them on virtual machines? Cohn says this is wishful thinking. "Because of cost pressures across the Street as a whole, firms are looking at ways to leverage their virtualization environments for any and all applications," he contends. "A given desk cannot afford to buy its own 70 machines and be off and rolling."