HPC APPLICATIONS: DESKSIDE HPC
high performance computing for small and medium companies,’ he says, adding that researchers want their own computational capacity to be able to run their own jobs. They want a way to quickly answer what Tanasescu calls ‘what if?’ questions without having to wait for time on a shared cluster.
Cloud is still blue-sky Practically, deskside HPC sits alongside cloud computing in offering small or medium-sized companies more control over their compute resources. There are many differences, as SGI’s Tanasescu explains: ‘First, if you don’t have a data centre, or if you don’t have a cluster, and you don’t have the expertise to operate a cluster, you can go to Amazon or SGI and you can go for cloud computing, buying or renting CPU cycles as required. Secondly you expand your PC or your personal workstation into something more complex that allows you to run more complex simulations,’ he says, referring to deskside HPC.
Cloud computing is, he says, not attractive for everybody: small and medium-sized companies in particular are very reluctant to put their knowledge (in terms of information and data) at risk. They can be paranoid, as data, information, and knowledge are the main assets they have. Maybe 10 per cent of these companies go to the cloud, but 90 per cent don’t… and for them, the capability to suddenly have the computing power to do what big companies do, on the desktop or under the desk, is very appealing. Cray’s Bolding adds another angle to this discussion, saying that while the cloud’s elasticity is one of its main selling points, many users are looking for a more specified environment: ‘Either they know how much resource they need, because they have to do a certain number of simulations in certain amount of time, or they’re doing development, in which case they’re testing very specific operating system implications and very specific configurations as they’re developing code, making sure that it all works properly. In either case, the cloud isn’t as controlled an environment for the type of user we’re looking at. In a cloud, you just say “I have this amount of work, please do it for me,” and the best cloud users don’t care whether it runs on
www.scientific-computing.com
an X86 or AMD processor, or an Nvidia accelerator; all they care about is that the simulation gets done. That’s a fine model, but we don’t believe that it’s something with which the developers and those running simulations will be comfortable.’
Mixing it up Both SGI’s and Cray’s systems are highly configurable, as are those offered by other providers. For many applications, nodes containing GPU-based accelerators can be included alongside the usual X86 nodes. A single GPGPU accelerator card added
to a standard workstation could, however, be considered a deskside HPC system. Sumit Gupta, head of management and marketing for the Tesla product line at Nvidia, explains that GPUs are particularly attractive for deskside HPC, as they offer impressive reduction of run time for certain codes while consuming relatively little power, and being cost-effective. Gupta gives the example of one particular Nvidia customer using Amber, the common molecular dynamics application, to carry out computational biochemistry. By switching from a 128-socket, X86-based
➤
Deskside HPC systems offer a range of different configurations, which can be tailored to meet the needs of the customer’s application
SCIENTIFIC COMPUTING WORLD OCTOBER/NOVEMBER 2010
33
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 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56