HPC: cloud computing


‘cloudburst’ users, but is ‘putting a very different spin on it’ by focusing on a high level of service. ‘We’ll deliver the output and the


computational power you need to deliver your business need, but because it’s [running on] our own manufactured equipment, we’re prepared to put a service level agreement around that, which will give you the certainty you need.’ Extreme Factory, again, offers ‘raw


compute power’, Carr says, with three levels of service: reserved capacity, custom offerings and a shared service. Not that virtualisation should be


dismissed out of hand: Roger Nolan, CTO of hosting software provider OnApp has seen virtualised clouds work well for HPC – it’s a matter of tailoring the offering, he says. ‘My personal view is that the advantages


of cloud outweigh the disadvantages, and nowadays, with modern hypervisors, the overhead isn’t huge,’ he says. Cloud providers simply need to adapt their cloud for the HPC clients, he says.


biology, chemistry, aren’t willing to become computer scientists fi rst just to be able to do their research. They want to focus on their discipline,’ he says. At Bull, too, ‘we prefer to maintain a level


of customer intimacy,’ says Carr. ‘Just giving a customer access to compute power doesn’t necessarily correlate to the output they’re looking for.’ This refl ects the experience of the


UK National Grid Service, which ran a pilot cloud service to see whether it was something users would be interested in. The main discovery, says research consultant Steve Thorn, of the University of Edinburgh, was that ‘it’s actually quite diffi cult for researchers to use’. A lot of the researchers who


experimented with the NGS cloud service had little IT experience themselves, and struggled to do anything useful. ‘At its most basic level, it’s quite diffi cult,


unless you’ve done that kind of thing before with physical hardware. And it’s made more complicated by the fact you


YOUNG RESEARCHERS, WHETHER IN PHYSICS, BIOLOGY,


CHEMISTRY, AREN’T WILLING TO BECOME COMPUTER SCIENTISTS FIRST JUST TO BE ABLE TO DO THEIR RESEARCH


‘I think most cloud providers today are


optimising for different things. Most of them are optimising for cost, and for availability generally – but it’s not that diffi cult to start optimising for speed, and I think more and more people are going to do that,’ Nolan says. One less cloud-like aspect of these non-


virtualised cloud services is that they all involve some interaction with the customer – there’s no ‘enter your credit card number’, Amazon-style self provisioning. ‘We’re building some more automation,


to allow customers to do things themselves more readily, but [for now] we use our expertise to assist them,’ says Wuischpard. Ease of use is a big issue, according to all


of the providers, and it’s still necessary to have some direct contact with each client to get them set up well. ‘We’ll assist them with making sure they


have the right libraries, their compiler’s set up – there’s a lot of little application things that we’re very familiar with,’ Wuischpard says. These new HPC customers aren’t as


technologically savvy as the traditional customers, often coming from a very different background, says Wuischpard. ‘Young researchers, whether in physics,


www.scientific-computing.com


don’t have physical access – you have to do it all remotely with some rather low level commands,’ Thorn says. That said, both Penguin and SGI say


they are working on ways to automate more of the service, as the current method of setting up clients ‘doesn’t scale well’, as Wuischpard says. The clients for these services range


from small companies – although these are generally still from the traditional engineering, scientifi c and manufacturing sectors – through to large life science clients running projects and university researchers getting frustrated with waiting for in-house systems. Penguin has also seen a number of PhD


students who have chosen to spend their grants on renting cloud services rather than buying equipment. Other customers include researchers from Japan, needing compute power after the tsunami, and the Icelandic weather service, running volcanic ash predictions. OCF’s fi rst client was race car design


company Lola Group. ‘We had a cluster that effectively broke,


at Christmas 2009, and we had to get a new one rather quickly,’ says the group’s senior aerodynamicist, Phil Tiller.


informatics for biofuels As the racing company’s supplier, OCF


was able to ‘lend’ the use of a remote cluster. ‘It had a few issues, but it certainly helped us out, and was a big help to us,’ Tiller says. Having had enquiries from other


customers, and seeing the success of its work with Lola Group, OCF went on to establish the enCORE service. Lola’s focus is on computational fl uid


dynamics, to improve the aerodynamics of its cars. The biggest issue they face in using an external, cloud-based computer is data transfer. ‘I send a load of input fi les to the


computer, get it to do the calculations, and then when it’s fi nished I want it to dump all the data onto disk and copy it back. The input data is relatively small, and would take about 10 minutes to copy across. But by the time it’s fi nished I’ve probably got in the order of 16 to 20 gigabytes of data that I need to get back, and if you’re trying to copy that over a standard broadband link it’s very slow,’ Tiller says. Lola Group has recently upgraded its


broadband link, giving ten times the speed, but in the past the company has been reduced to couriering data back and forth – hardly the ideal of cloud computing. The racing company does have its own, in-house computing system, Tiller stresses, but uses the OCF cloud offer as and when it needs it. ‘We’re a motor company, but we’re not


F1. We’re able to justify capital expenditure to a certain extent, but you end up buying quite a modest machine. [So] if all of a sudden we get a project with short time scales or a big design project that needs lots of intensive work, then we need extra capacity. It’s great – additional capacity that we can just turn to. It’s like having an extra computer next door,’ he says. What if you didn’t have to transfer data


at all, yet could still use ‘cloud’ HPC? Private clouds are beginning to appear here, just as in standard cloud computing. While most of Bull’s Extreme Factory


users are either small businesses or ‘cloudburst’ clients, Andrew Carr says a number of clients are looking at ‘an internal, private-cloud based HPC. Because what they typically have is a number of clusters that sit within different lines in the organisation. And what they’re trying to do is take that back into the centre, and deliver the HPC compute power back out to the business lines.’ Likewise, SGI has already set up a private


cloud for a client, an oil and gas customer running seismic processing worldwide.


DECEMBER 2011/JANUARY 2012 21


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