When you consider technologies that make corporate IT more efficient by improving utilization of computing resources, VMware and its x86 virtualization software may be what that comes to mind.
When you consider technologies that make corporate IT more efficient by improving utilization of computing resources, VMware and its x86 virtualization software may be the first that comes to mind.
But there's a newer, much smaller vendor on the scene that has attracted numerous Fortune 500 customers with grid and cloud management technology that reuses wasted compute cycles and automates the process of building HPC clusters.
Cycle Computing, founded five years ago primarily to help customers use Condor, a workload management system that can run compute-intensive jobs on idle PCs, has grown to manage numerous job scheduling systems, help customers reuse wasted capacity in VMware installations, and support cloud services such as Amazon's Elastic Compute Cloud.
At the recent SC10 supercomputing conference in New Orleans, founder and CEO Jason Stowe ticked off an impressive list of customers, including JP Morgan Chase, Pacific Life, Lockheed Martin, Eli Lilly, Pfizer, Bank of America, Electronic Arts, Johnson & Johnson and ConocoPhillips. JP Morgan Chase, for example, uses Cycle Computing technology to run large calculations on desktop cycles that would otherwise go to waste, according to Stowe.
Cycle is bootstrapped with no external investors, and pulls in "a couple million dollars" in revenue from fewer than 100 customers, according to Stowe. Cycle has fewer than 20 employees and has been growing revenue 50% a year for the last four years, he says.
Just as VMware borrowed technology from IBM's mainframe and applied it to the modern data center composed of x86 servers, Cycle Computing has taken strategies based on decades-old research and adapted them to the age of cloud computing and virtualization.
The idea of harnessing unused cycles from desktops to build a computing cluster has "been around for 20 years at this point," IDC analyst Steve Conway says. (Condor itself was started nearly 25 years ago). Cycle Computing has succeeded because it automates the difficult tasks that have to be performed by IT administrators, Conway notes, saying "Everyone's looking for ease of use."
But Conway says what he's most excited about regarding Cycle Computing is the vendor's long-term plan for making clustering technology viable in cloud computing services like Amazon EC2. While Conway believes the technology is not quite there today, over the next few years "it could be possible to replicate in public clouds the corporate, on-premise computing environment, with all the policies, directives, and all the rest of that intact. It truly can be an elastic extension of the corporate environment."
The CycleCloud product, built upon Amazon Web Services, lets customers create high-performance computing clusters that automatically turn nodes on and off as jobs are started and completed.
In about 10 or 15 minutes, the CycleCloud software "spins up an entire HPC environment inside Amazon infrastructure," with shared file systems, a terabyte of storage, 10 Gigabit per second interconnect and Intel Nehalem chips, Stowe says.
While CycleCloud manages external resources, the vendor's CycleServer product targets the internal data center, providing a Web-based interface for managing HPC clusters based on Condor, SGE, Torque and Hadoop. Cycle's technology is operating system-agnostic, working with both Linux and Windows.
One of the more surprising use cases is built around data centers making heavy use of the VMware hypervisor. While VMware's technology drives up utilization by allowing multiple operating system images and applications to run on the same machine, these servers are still underutilized, particularly at night, according to Stowe.
"VMware is about footprint consolidation. It doesn't change the nature of peak vs. median usage," Stowe says. "You have these servers that are used incredibly heavily at certain points, and not at all at other points. Virtualization has no impact on that."
Improving utilization of VMware servers is an area where Cycle Computing is doing an increasing amount of work, and has "a pretty broad patent application around it," he says. "If you get 20% to 40% out of your VMware environment, that's a pretty awesome job you've done in pulling out CPU utilization," he says. "But you're still leaving 60% capacity on the floor every evening. We can turn around and make it 90-plus."
While Cycle's products have generally been targeted at technical computing scenarios, theoretically any business with unused computing capacity could benefit from them. Cycle's competitors include SGI and WiPro, and Amazon and Microsoft, Conway says. Amazon, for example, is now offering cluster instances based on graphics processing units, and Microsoft's Windows HPC Server is now harnessing idle Windows 7 desktops.
Conway says Cycle Computing tends to be more expensive than its competitors because its products are so specialized, but he has confidence in the young company's future.
"What I like about Cycle in particular is the vision," Conway says. "They really have a very strong and detailed vision of how cloud computing is going to evolve in the next two to five years."
Follow Jon Brodkin on Twitter: www.twitter.com/jbrodkin