Cloud in HPC
Gillian Law explores how the HPC industry has taken up the benefi ts of cloud computing
C
loud has been late to take off in HPC. In a world where you want to dredge every last ounce of power out of your computer system in order to run as
fast as you can, the overheads brought in by a virtualised resource just seemed a deal-breaker. And yet there’s such an obvious market
there, in smaller businesses that can’t afford to buy and run their own cluster, in short- term projects run by bigger institutions, and in ‘cloudbursting’, when established HPC set ups get temporarily overwhelmed. With all that business out there, it was
only a matter of time until suppliers hit upon a solution: non-virtualised HPC ‘cloud’ offerings that let users buy the processing power they need, but can run at the speed and effi ciency of true HPC. At Penguin Computing, CEO Charles
Wuischpard came up with the idea for Penguin on Demand two years ago, and is seeing it really begin to take off. ‘We’d see people who didn’t have the skills to run things in-house, and budgets
20 SCIENTIFIC COMPUTING WORLD
were a problem, and all sorts of pain points. So we started. And the difference between the system that we put in place and what Amazon offers, and the way that cloud’s traditionally designed, is that it’s not a virtualised environment. It’s a supercomputer, a Linux cluster in the cloud. It’s Infi niband-connected, it has parallel fi le system access, we’ve got a portion with GPU – so there’s nothing missing from it. In fact, it’s a lot more capable in many cases than what people would buy themselves,’ he says. Penguin had to build a billing system and
a scheduler, along with some extra security measures and ways of moving data quickly, but can now support a large number of users, Wuischpard says. SGI has set up a similar offer, with its
Cyclone service, as has Bull with Extreme Factory and OCF with enCORE – and no doubt many others. ‘Interest in using cloud resources for HPC
has defi nitely gained in popularity,’ says Bill Mannel, VP of products at SGI. In between the small group of large
organisations with their own HPC facilities, and the small users ‘confi ned to the capabilities of their PC or laptop’, there’s a group that has been identifi ed as the ‘missing middle’, Mannel says. ‘All these people could be using a lot
more HPC, but they don’t have the money; they don’t have the skills, the access or the knowledge to do that. Now they can just go and get the resources on the web,’ he says. Cyclone ‘is virtualised to the extent that
the customer doesn’t know what specifi c machine they’re running on’, but doesn’t use any virtualisation software, Mannel says. Likewise, OCF offers a ‘bare metal’
service on its enCORE service. ‘It’s a non-virtualised environment, so you’re running directly on the processor, on the node, as opposed to a virtualised machine. That has a cost in terms of the numbers of users you can put on there at one time,’ says cloud business development manager Jerry Dixon, but you avoid the ‘signifi cant reduction in performance’ that virtualisation would bring. Andrew Carr, Bull’s sales and marketing
director, says that Extreme Factory is cloud-like in the way it offers services on- demand, both to small clients and to larger
www.scientific-computing.com
Kasza/
Shutterstick.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40