This post starts as a throwback to the utility and grid computing applications that used to dominate headlines.
The high-performance grid in question is Purdue University’s DiaGrid, which aggregates the idle compute power of 28,000 processors at the university and on campuses in Indiana, Kentucky and Wisconsin. What initially started as a project mainly focused on effective resource utilization has, over time, has become a potential method for harvesting energy across the connected systems, says John Campbell, associate vice president at Purdue’s Rosen Center for Advanced Computing. What makes this possible is the Condor and CycleServer management tools from Cycle Computing.
The directive is pretty simple at the university, which is trying to eke every available dollar out of the workstations and academic computers across the high-performance computing cluster, which are typically idle between midnight and 7 a.m. “Either join Condor or turn off your machine at night to conserve power,” Campbell says. Eventually, they won’t won’t have to make that decision: the software will automate a shutdown of idle machines.
To get a sense of the impact that DiaGrid has had on the Purdue IT budget, Campbell notes that if the university was forced to replace the computing cycles that the grid coordinates, it would have to spend roughly $3 million in new hardware, not to mention all the power required to run those new systems. The power discussion has grown louder in the past two years, Campbell says, as the campus seeks to get the most utilization out of every watt of power consumed. That conversation has inspired other universities to join the DiaGrid project.
Jason Stowe, CEO and founder of Cycle Computing, says many of the company’s clients — which include the likes of JP Morgan Chase, Lockheed Martin, Eli Lilly and Pfizer — are looking at how high-performance technical computing clusters can play a role in managing power management costs. While the power savings potentially might not be enormous, grid utility applications can play a key role in making sure a company’s existing power draw is used as effectively as possible.
“Rather than buying new machines and more data center space, these technologies can help them make better use of what they have and let them power down nodes that are no longer in use,” Stowe says.
