Cloud computing is everywhere. Supercomputers on the other hand, are not. Amazon thought it was time to fill the gap and has made a supercomputer in the cloud. A supercomputer isn’t just one, huge computer that you turn on and off with a power button. It’s a combination of thousands of processors and sticks of memory (all in different machines, networked together) that are all harnessed into one purpose, as though it was one computer that had all of these resources available to it.
Amazon first built this infrastructure in 2006 but only made it into Top500.org’s annual rundown of the fastest supercomputers on Earth beginning in 2010. However, they hovered near the bottom of the list. But for 2011, Amazon upped their game and their cloud-facilitated supercomputer was dubbed the 42nd fastest supercomputer in the world.
When is this kind of power necessary, you might ask? Research facilities use supercomputers all the time to crunch billions (usually more) of calculations at once to speed through combinations of problems or experiment variables. Modeling, simulating, and forecasting data is a supercomputer’s main purpose.
So how does a supercomputer operate in the cloud? Instead of having one hundred racks of dedicated servers networked together to form the conglomerate supercomputer, Amazon uses any combination of available servers in the cloud at any one time to make up the force behind their supercomputer. Probably the most impressive part about Amazon achieving the 42nd fastest supercomputer in the world is that it’s powered by their standard EC2 web services.
While most supercomputers have an entire data center at their disposal, Amazon is calmly carrying on with its other cloud services while this supercomputer is plugging away amidst it all. Also, the addition of a virtual layer on top of all of the hardware that makes up the behemoth processing power is a feature unique to the cloud supercomputer and not to other dedicated supercomputers.
Critics point out that not all supercomputer-specific applications are equipped to run on the cloud. Jack Dongarra oversees the Top500 supercomputer annual list, and is also a professor at the University of Tennessee. He says that some of the top 25 supercomputers within his Top500 list are comprised of specialized hardware, designed specifically to run very specialized software. Without a lot of redesigning, that kind of software wouldn’t see the same benchmark results from a supercomputer in the cloud, though it would run to some degree. Dongarra praises what Amazon has done for people that don’t need the fastest supercomputer in the world (which, for the record, completes 10 quadrillion calculations per second), just a pretty fast supercomputer (Amazon’s completes 240 trillion calculations per second).
Some supercomputing companies realize that a virtual element to access their resources would be incredibly valuable. Penguin Computing offers Penguin-on-Demand (POD), which lets users and companies who would like to use their supercomputer access it online. The primary difference between their new service and Amazon’s cloud is that Penguin technicians can point directly to the machines that a user is employing for their research, whereas Amazon is using a hodgepodge of machines available in a world-wide network of data centers. One Penguin customer, Earthmine, prefers POD to working in the cloud because they know they’ll receive a deeper level of support (i.e. monitoring of jobs and processes on the hardware) than Amazon will provide.
But, companies have recognized that shortcoming from Amazon’s supercomputer and have stepped in to fill the need and make some money. Cycle Computing helps companies utilize processor cores on top of the cloud and provides the support that POD offers. Recently, a pharmaceutical company needed a 30,000 core supercomputer and Cycle stepped in to set it up in Amazon’s cloud.
One of the most publicized details of that interaction was the cost, which is another huge benefit to Amazon’s supercomputer. A 30,000 core supercomputer made of dedicated servers might cost millions of dollars an hour. An hour with Amazon’s 30,000 core supercomputer? $1,279. It might not be the fastest, but it’s in the top 50, and for that price, it’s nothing to balk at.