Feature - Enter the age of computer merchants
Modern researchers are not short of ambition. Physicists at the Large Hadron Collider at CERN, for example, will need to catch and sort through an estimated 15 petabytes of data each year, the equivalent of 20,000 years of music on an MP3 player.
Collaborations like WISDOM, a global initiative for discovering new medicines for neglected and emerging diseases, test tens of millions of chemicals in computer models.
This is an opportunity for companies like Digital Ribbon in the United States, which seeks to be a kind of clearing house for computational resources. They call their model a “service registry,” connecting resource consumers with the right resource providers. It could transform the way users run jobs on clouds or grids.
“Computing is moving towards becoming a commodity,” says Erik Weaver, Digital Ribbon’s CEO. “However it is much more complex than the oft-cited example of electricity.”
Computing consumers care about more than power, they also care about bandwidth, interconnectivity (how fast individual processing cores can talk to each other) and memory. It is important to have not just enough resources, but also the right kind. As Weaver puts it: “If you are making an apple pie, you might want to use Granny Smith, not Fuji apples.”
Having more options could be a great advantage to users. “We are always looking for ways to link user communities, applications and resource providers,” says Bob Jones, project director of EGEE.
As a step towards smoothing the way for EGEE and Digital Ribbon collaboration, the WISDOM collaboration ran a test run early this summer, using gLite middleware on Digital Ribbon resources.
Quantity, quality . . . commodity?
“This successful test with WISDOM shows that applications within EGEE can run well on Digital Ribbon resources,” says Jones. “Now we can see if other user communities might benefit.”
Data used for the test consisted of 750 chemical dockings—potential candidates for diabetes drugs. In about 12 hours they ran calculations corresponding to 55 days on a single processing unit.
“While this was small for us in comparison to previous sets of calculations,” says Jean Salzemann, computing researcher with the WISDOM collaboration, “it tested new ways of sending jobs. When we are preparing future data challenges, we could consider splitting the data load between EGEE and Digital Ribbon resources.”
In their last data challenge, WISDOM tested 4 million potential drug candidates using 5,000 processing cores. With these they were able to run in two months what it would have taken a single CPU four centuries to complete. If they had 10,000 cores, they could have done the calculations in one month.
“We test millions of potential drugs in our data challenges,” says Salzemann, “ideally we would have access to one core per potential drug or ‘ligand.’ With this we would have the complete results in about an hour.”
—Danielle Venton, EGEE
Digital Ribbon's service registry covers a multitude of platforms, processors, architectures, offering resources to public and private organizations. More information about using gLite in business environments is available online.