Thursday, September 20, 2012

Build your own supercomputer

Many image processing tasks are computationally intensive. As such, system integrators are always on the lookout for any means that will help them to accelerate their application software.

One way to do this is to determine whether an application could be optimized -- either by hand or by using optimization tools such as Vector Fabrics' (Eindhoven, The Netherlands) Pareon -- to enable it to take advantage of the many processing cores that are in the latest microprocessors from AMD and Intel.

If an application can be considered to be easily separated into a number of parallel tasks -- such as those known in the industry as "embarrassingly parallel problems" -- then the only limitation the systems integrator has is how to source enough inexpensive processors to perform the task.

Fortunately, since the advent of the GPU, cores are plentiful. As such, many engineers are harnessing the power of games engines such as GE Force’s GTX 470 -- which sports no less than 448 CUDA cores and 1GByte of memory -- to vastly accelerate their image processing applications.

Now in a few cases where engineers really need to harness even more hardware power, they have only one alternative -- build it themselves. That, indeed, is exactly what engineers at the Air Force Research Laboratory (Rome, NY, USA) have done.

Their massive supercomputer -- which was developed for the Air Force for image processing tasks -- is ranked as one of the fortieth fastest computers in the world. Yet, believe it or not, it has been constructed by wiring together no less than 1,700 off-the-shelf PlayStation 3 gaming consoles!


Now if you are anything like me, you are probably wondering how you might be able to design such a beast yourself, while doing so without shelling out an inordinate sum of money to buy so many Sony games consoles.

If you do, you might like to check out the web page of Professor Simon Cox from the University of Southampton (Southampton, UK), who, together with a team of computer scientists at the university (and his six year old son James) has built a supercomputer out of Raspberry Pi's, a rats nest of cables and an awful lot of Lego.

"As soon as we were able to source sufficient Raspberry Pi computers we wanted to see if it was possible to link them together into a supercomputer. We installed and built all of the necessary software on the Pi starting from a standard Debian Wheezy system image," says Professor Cox.

The machine, named "Iridis-Pi" after the university's Iridis supercomputer, runs off a single 13A mains socket and uses a Message Passing Interface to enable the processing nodes to communicate over Ethernet. The system has a total of 64 processors and 1Tb of memory (16GByte SD cards for each Raspberry Pi).

Now I'm not about to claim that this supercomputer is going to rank up there with the PlayStation-based system built for the US Air Force, but it certainly would be a fun project to build and experiment on. And at a price of under $4000, who wouldn't want to give it a go?

Fortunately, for those interested in doing so, the learned Professor has published a step-by-step guide so you can build your own supercomputer Raspberry Pi supercomputer without too much effort.

The Southampton team wants to see the low-cost supercomputer used to enable students to tackle complex engineering and scientific challenges. Maybe the system isn't really the most cost effective way to do that, but it certainly is inspirational.

Editor's note:  PA Consulting Group and the Raspberry Pi Foundation have teamed up to challenge schoolchildren, students and computer programmers to develop a useful application using a Raspberry Pi that will make the world a better place. I'm sure they would welcome ideas from the imaging community! Details on the competition can be found here.


No comments:

Post a Comment