This page contains a Flash digital edition of a book.

More complex clusters mean more scope for integrators – provided they know what their customers

need, as Tom Wilkie discovers


ntegrators ‘make complex solutions out of simple components,’ jokes Julian Fielden, managing director of UK-based HPC integrator OCF. Yet as the technology

of high-performance computing becomes more complex – with the arrival of GPU accelerators, Intel ‘co-processors, and so on – and as customers’ needs become more complex – with markedly different types of jobs being run on ‘private clouds’ – is there really scope for small to medium-sized integrators to serve the market? Will they not inevitably be out- stripped by the major players, the IBMs, Dells, HPs and Crays of this world? According to Eva Cherry, president and

CEO of Silicon Mechanics: ‘Te wind is changing, in a good but challenging way. Te complexity and the technology that drives cloud, virtualisation, and big data, really forces organisations to look at re-architecting their environment.’ Within limited budgets, there is a need for the system to be flexible, which means a shiſt to commodity hardware, she continued. ‘Our thinking was: “What if we were able to offer a different type of experience to our customers? One that doesn’t just focus on hardware components, or even how to integrate them, but one where value is being added from the moment that someone gets in touch with Silicon Mechanics?” We made it our business to create an individualised customer experience for every customer who comes to us.’ Her analysis of the state of HPC and the role

of the integrator chimes precisely with Fielden’s on the other side of the Atlantic: ‘My view is that integrators are vital, because there is no


one Tier 1 that supplies the whole solution.’ According to Fielden: ‘Te solutions are now far more complex than they were. Tere is a requirement to balance the compute and the I/O to make the system efficient, and to get intelligent scheduling to ensure that time is not wasted. Tere isn’t a Tier1 that does all that.’ David Power, head of HPC at UK-based

Boston, agrees with the growth in complexity: ‘You’re looking at a complex matrix of permutations and components within HPC nowadays’. Boston might originally have been considered more of a component distributor but, he continued: ‘Over the years we evolved to reach into the solution and HPC space. We are today capable of delivering full turnkey HPC solutions for our end-customers. From the hardware, to the soſtware stack, to the application integration, we do all of those steps in house and can deliver and work with other partners to deliver HPC solutions for different markets.’



But it is not just the technology that is

becoming more complex. Te purposes to which users put their systems are also evolving, according to Fielden. ‘Systems are now becoming multi-purpose,’ he said. ‘Ten years ago, you’d buy a machine that would run one code, maybe. It was easy to put together and design. Now, where systems are required to be production machines serving internal customers – an internal cloud if you will – there’s a far greater requirement for things other than the “tin” to make it work efficiently. Tat’s why there needs to be an integrator, as opposed to a reseller. Te thing needs to be designed for purpose.’ Power from Boston agrees with this trend,

A fair wind of change for

especially among university-based customers: ‘Within the academic space, HPC is a central tool for a wide range of diverse academics to fulfil their scientific roles. A decade ago, a university might have had ten different clusters, hidden around the place in broom cupboards. Tere has been a drive to centralise the resource, where you will have Linux and compilers experts working with scientists to run their code. But this centralisation also introduces complications. Computational chemistry people might have the need for much larger globally addressable memory space; whereas if you look at theoretical physics, those guys might be running parametric sweeps and want more and more cores. Tis is where getting that balance becomes a challenge.’ On the other hand, he said: ‘Commercial users are more finely tuned around single applications, so with manufacturing we might be trying to optimise CFD-type applications. We have to tweak and amend our standard platforms, either to be highly tuned for specific requirements, or to try and make them well balanced for a wider range of requirements.’ Te same theme permeates Eva Cherry’s

discussion: ‘Customers want their lives to be easier and no negative surprises. Tey expect us to get it right and consistently so. Te bottom line is the ability of an integrator to create a customer experience. How do we differentiate ourselves? It’s about the customer experience that we have to create.’ Te trend in academic research in the USA

however, does not mirror that in the UK and to some extent the rest of Europe. According to Cherry, rather than centralised facilities buying the supercomputer, for Silicon Mechanics: ‘Te majority of our sales in research computing are grant-based or start-up fund based. What a lot of universities do is create a central purchasing function with an approved vendor list, so it is easier for the researchers to buy what they need. But researchers are typically allowed to buy from whatever vendor they wish to work with. For integrators, it’s important to be on both that list of approved vendors and

@scwmagazine l

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