(Super)computing On the Cloud

January 9, 2012 Off By David
Grazed from CloudTweaks.  Author:  Sourya Biswas.

$1279 an hour – seems a lot to hire a computer, right? What if it’s a supercomputer capable of performing 240 trillion calculations per second, or 240 teraflops (a flop is the acronym for floating point operations per second, the universal measure of a computer’s performance)? This is the performance promised by the latest innovation from the Amazon stable – the supercomputer on the cloud.

This cloud supercomputer runs on Amazon Web Services Elastic Compute Cloud (EC2) and features Intel’s Xeon 8C 2.60 GHz processor with 10G Ethernet interconnects providing 65,968 GB of capacity and 17,024 cores. As per the latest supercomputer rankings, this cloud supercomputer ranks as the 42nd fastest in the world. Considering that it’s the only one on the cloud and built with existing non-specialized equipment even as the company continued normal operations, that’s extremely impressive…

Now, let’s put this into perspective. The typical high-end laptop runs at around 7 gigaflops, or some 40,000 times slower than the Amazon supercomputer. Admittedly, there are several supercomputers which run much, much faster; however, they also cost much, much more. For example, the fastest supercomputer in the world today, the Fujitsu K at the RIKEN Advanced Institute for Computational Science campus in Japan, has surpassed the 10 petaflop barrier, equivalent to 40 times the speed of the cloud supercomputer. However, it also cost $1.2 billion to develop (See: Japan creates world’s fastest supercomputer which is as quick as one MILLION desktop PCs ). Considering that an Amazon client can simply hire the cloud supercomputer at $1279 per hour and the underlying hardware cost Amazon around $20 million, it seems like a win-win situation for everyone.

The ability to pay as per usage without having to spend on capital expenses is one of the major advantages of cloud computing (See: How Cloud Computing Can Save You Money). Considering that few organizations will require the services of a supercomputer for any sustained period of time, this advantage is even more pronounced here.

This cloud supercomputer has already had a successful run. According to Cadre Metz of Wired magazine,  Cycle Computing had set it up this fall for a pharmaceutical company. “If you wanted to spin up a ten or twenty thousand [processor] core cluster, you could do it with a single mouse click,” said Jason Stowe, the CEO of Cycle Computing. From web-based email to supercomputers, cloud computing has come a long way!