Feeds

Google donates a billion cores to boffins

Hundreds of teraflops for the taking

Beginner's guide to SSL certificates

Universities, government labs, and sometimes IT vendors donate their excess supercomputing capacity through grants to academics to help advance various sciences. Now Google is letting boffins loose on its systems.

In a blog post, Alfred Spector, vice president of research and special initiatives at the Chocolate Factory, said that Google had created an academic research grant program called the Google Exacycle for Visiting Faculty, which will donate one billion core-hours to science.

Google says that this level of computing is orders of magnitude more computing than most academics can get their hands on, no matter how big the endowment is at the university or how much research they do for government.

Google is not just giving away compute cycle on its massive server clusters to get a tax write-off on unused capacity, but to blind us with science. "Google Exacycle for Visiting Faculty is not a conventional grant program," the company claims. "We aim to stimulate advances in science and engineering research by supporting the computational needs of projects that push boundaries and reach for remarkable breakthroughs."

Google is not giving all of the billion core-hours to one lucky researcher. The plan is to solicit distinguished researchers and postdoctoral scholars from all over the globe and award them grants for jobs that can chew through at least 100 million core-hours.

Those who win the grants will do their work from Google offices and sign an employee agreement with Google for the term of the simulation. You have to pay your own travel, lodging, and living expenses while the simulations run.

The company says that large-scale genomics and protein folding simulations are the kinds of jobs it expects to most benefit from such a large number of cores to frolic upon; embarrassingly parallel jobs will do best, and "pleasantly parallel" jobs (yes that is a technical term) will work.

"The higher the CPU to I/O rate, the better the match with the system," Google says, and jobs that have minimal communication between nodes will do best. (Sounds like Gigabit Ethernet to me.) Your program has to be coded in C/C++ and compiled via Google's Native Client SDK, its tweak of the open source GNU C++ toolbox. Sorry, no Fortran or Java apps need apply. Researchers have until May 31 to apply for the capacity.

Looking ahead, Spector says that Google is thinking of extending CPU capacity grants to businesses in various industries, including biotech, financial services, manufacturing, and energy. Spector did not say that these grants would be free – he didn't say Google would charge for them, but it makes sense that it would – and is soliciting ideas from industry now on what jobs companies might want to run.

So just how much is a billion core-hours in terms of HPC capacity?

The largest cluster of Xeon machines in the world not using a proprietary interconnect of some kind is the Pleiades supercomputer at NASA's Ames Research Center. It uses Intel's old quad-core Xeon 5400 processors from two generations ago in two-socket machines; the cluster has 81,920 cores running at 2.93 GHz and links the servers together with an InfiniBand network.

Those chips can issue four floating point instructions per clock cycle per core, which works out to over 960 teraflops of aggregate peak number-crunching power. (On the Linpack floating point test, the Pleiades machine delivers 772.7 teraflops of actual performance.) If you ran the Pleiades machine flat out for a full year, you are talking about 718 million core-hours.

A grant of 100 million core-hours is around 11,408 Xeon cores running for a full year, and with modern six-core Xeon 5600 processors, you are talking about Google giving 950 server nodes. (Obviously, if you want to run that job in three months instead of 12, you have to quadruple the server node count.)

Google has millions of servers, so this is a tiny fraction of what the search giant has running in its 36 data centers. Depending on how fast you want to burn those cores, the virtual HPC cluster that Google will grant you could be rated from one to several hundred teraflops.

So the Google grants may be a tiny piece of Mountain View's capacity, but the capacity Google is putting up for grabs is a lot more than most researchers can get their hands on.

You can find out more about the the Google Exacycle for Visiting Faculty here. ®

Security for virtualized datacentres

More from The Register

next story
It's Big, it's Blue... it's simply FABLESS! IBM's chip-free future
Or why the reversal of globalisation ain't gonna 'appen
'Hmm, why CAN'T I run a water pipe through that rack of media servers?'
Leaving Las Vegas for Armenia kludging and Dubai dune bashing
Microsoft and Dell’s cloud in a box: Instant Azure for the data centre
A less painful way to run Microsoft’s private cloud
Facebook slurps 'paste sites' for STOLEN passwords, sprinkles on hash and salt
Zuck's ad empire DOESN'T see details in plain text. Phew!
CAGE MATCH: Microsoft, Dell open co-located bit barns in Oz
Whole new species of XaaS spawning in the antipodes
AWS pulls desktop-as-a-service from the PC
Support for PCoIP protocol means zero clients can run cloudy desktops
prev story

Whitepapers

Choosing cloud Backup services
Demystify how you can address your data protection needs in your small- to medium-sized business and select the best online backup service to meet your needs.
Forging a new future with identity relationship management
Learn about ForgeRock's next generation IRM platform and how it is designed to empower CEOS's and enterprises to engage with consumers.
Security for virtualized datacentres
Legacy security solutions are inefficient due to the architectural differences between physical and virtual environments.
Reg Reader Research: SaaS based Email and Office Productivity Tools
Read this Reg reader report which provides advice and guidance for SMBs towards the use of SaaS based email and Office productivity tools.
Storage capacity and performance optimization at Mizuno USA
Mizuno USA turn to Tegile storage technology to solve both their SAN and backup issues.