You are here

Speaking of linux clusters...Roadrunner is /fast/

I was happy to learn on Monday that the Petaflop barrier has been broken. IBM's Roadrunner supercomputer achieved this feat with commodity hardware and open source software (including Red Hat's Enterprise Linux).

Achieving this level of performance is something that many in the research community have dreamed of for a long time. When it was reported in the press this week, many reporters compared it to breaking the 4-minute mile, a limitation we now know was arbitrary, but at the time seemed somehow impossible. And, happily, that characterization helped the non-technical public share the excitement and triumph of the accomplishment, with all manner of news outlets telling the story for days.

For me, this story proves that the logic of open source software is now fully being demonstrated by concrete results. Why should people spend their hard-won capital on somebody else's proprietary technologies when commodity technologies offer both a faster intrinsic innovation curve (Moore's Law, Fiber Law, and Disk Law are all alive and well)? Why should people paint their designs into a proprietary corner when open standards and open source software offers more choice and more control over interface and implementation decisions? The Roadrunner implementation shows just how well an open system lowers barriers (to science and discovery) rather than creating them (and forcing scientists to guess not only the target of proper inquiry, but also its path).

And yet it should be noted that while we have now achieved the Petaflop (1.026 Petaflops to be precise), that is only a small fraction of what people estimate the human brain to be capable of doing. The best estimate I've heard (which, though ten years old, still seems supported by the neuroscience community) is that the brain achieves the equivalent of 40 Petaflops. Moreover it does so while consuming an estimated 20 Watts. With 296 racks spanning 12,000 square feet of datacenter space, I suspect that not only is there still a 40:1 advantage in terms of human:computer CPU power, but more than a 40:1 advantage in power efficiency, and for that reason we should honor and respect the computing potential of all who choose to keep their minds open, learning, and engaged.