HPC 2012 Processing He continues: ‘I’m not suggesting for one Oak Ridge National
Laboratory’s Titan supercomputer combines CPUs and GPUs
be difficult,’ comments Jerry Dixon, business development manager at OCF. He emphasises that making a capital investment is fraught with danger for the uninitiated – high-performance equipment ages quickly – and that maintaining the investment in new equipment can be a burden. Tere is a solution, however, enabling organisations to access compute power that, whether through a lack of resources, financing or specialist skills, would otherwise be unattainable in-house. Tis solution, as Dixon explains, is the cloud:
businesses need pay only for the compute power that they actually use, they can use the service with confidence that all the equipment is fully up to date, and concerns over energy, cooling and space are transferred to the supplier. ‘However,’ he warns, ‘the reality of high- performance computing in the cloud is more grounded than many suppliers would have you believe. A complete plug-and-play cloud service for high-performance computing is just not possible, it possibly never will be, and possibly nor should it be.’
minute that high-performance computing in the cloud should be avoided, far from it. For smaller business particularly, the benefits of cloud far outweigh the technical “hurdles” of using such a service. However, I am suggesting that users should think carefully about their own technical capability to decide whether they can work autonomously, or if they require technical support from a supplier – to run trials of the application, to obtain compilers if needed, and to courier data to and from the sites. Tey should then make a decision about how to access high-performance computing in the cloud.’ Dixon adds that he believes that most
businesses and users will fall into the latter category where technical support is essential. With that in mind, his message is simple: ‘pick a supplier, partner up and let them guide you through the process.’ Eurotech’s Giovanni Mattiussi has a final,
and crucial, message of his own: ‘Beyond a technological point of view, it’s important for Europe to unite behind high-performance computing.’ He believes that with the vast amount of developments occurring in the US, Japan and China, Europe needs to gain not only a competitive edge within the supply chain, but to gain access to new technologies as soon as they are developed. ‘We do have the technology,’ he says, ‘but are missing the structure. With that in mind, a platform for HPC in Europe is being built, but it’s at the very early stages and there needs to be a clearer understanding of what we hope to achieve. If trends continue in this direction, then hopefully the end result will be a better structural alignment between what HPC technologies can do and the needs of users within the public and private sectors.’ l
Further information
AccelerEyes
www.accelereyes.com
Acceleware
www.acceleware.com
Allinea
www.allinea.com
AMD
www.amd.com
ARM
www.arm.com
Atipa Technologies
www.atipa.com
Calxeda
www.calxeda.com
Caps Entreprise
www.caps-entreprise. com
Comsol
www.comsol.com
Convey Computer
www.conveycomputer. com
EM Photonics
www.emphotonics.com
Eurotech
www.eurotech.com
Fluidyna
www.fluidyna.de
Freescale
www.freescale.com
Fusion-io
www.fusion-io.com
Gompute
www.gompute.com
Intel
www.intel.com
Maxeler Technologies
www.maxeler.com
Microsoft
www.microsoft.com
NAG
www.nag.co.uk
Nallatech
www.nallatech.com
Numascale
www.numascale.com
Nvidia
www.nvidia.com
OCF
www.ocf.co.uk
Oracle
www.oracle.com
Oxalya
www.oxalya.com
PNY
www.pny.eu
Scale MP
www.scalemp.com
Supermicro
www.supermicro.nl
Symmetric computing
www.symmetric
computing.com
SysFera
www.sysfera.com
Texas Instruments
www.ti.com
The Mathworks
www.mathworks.com
The Portland Group
www.pgroup.com
Univa
www.univa.com
Vcodyne
www.vcodyne.com
Whamcloud
www.whamcloud.com
Debugging for HPC applications David Lecomber, CTO at Allinea Software, discusses the role of debuggers
Soſtware development has rarely occupied such a pivotal role in the success of so many HPC projects as it does today. Many-core processors and accelerators, such as Nvidia’s Cuda GPUs and Intel’s Xeon Phi processors, coupled with the sheer size of systems today have each added challenges to getting soſtware to achieve maximum performance from the hardware resources. Developers are adapting scientific applications by restructuring and rewriting major components – or are developing new applications from scratch. Whenever soſtware is developed, the
soſtware bug is on-hand to spoil the day. Traditionally, many developers have been reluctant converts to debuggers – relying on print statements and debug logs instead – but times have definitely changed. Tose old print and debug log techniques just don’t cut it today – the scale and complexity of many thousand processes or threads interacting with each other make output unmanageable. Debugging tools are the first port of
call in today’s complex HPC environments when trouble strikes. Successful debugging of complex applications and environments
involves two critical parts. Te first is having the information necessary to understand what is happening – seeing the application state in the most effective way possible. Te second is to be able to debug regardless of the size or architecture of the machine. With Allinea Soſtware’s parallel debugger,
Allinea DDT, developers can see the variables and state of every process and thread within an application – including those on accelerators. Te guesswork of print statements and its iterative debugging is gone because the entire state of the application is on hand interactively.
17
Oak Ridge National Laboratory
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