The University of Oxford’s new Arcus Phase B HPC cluster, integrated by OCF, includes Lenovo NeXtScale servers with Intel Haswell CPUs connected by 40GB Infiniband to an existing Panasas storage system.
Dr. Andrew Richards, head of Advanced Research Computing (ARC), the University of Oxford’s HPC facility, talks high performance computing with Information Age.
What is your role in getting more students involved in HPC and research?
ARC provides hardware, software and, most importantly, user support. The user support takes the form of direct one-to-one help, consultancy and training courses. That’s our value proposition and one that OCF has really helped us to achieve
For many researchers we are their first step into using any form of HPC. DPhil students are our normal entry-level users, though we do occasionally support undergraduate and masters training courses that focus on the use of one particular application.
To facilitate this approach, ARC operates as a partnership between the central IT Services department and the Oxford e-Research Centre. The partnership with the OeRC helps the ARC service to remain agile and to respond quickly to researchers changing requirements in supporting research.
Whilst our main HPC services are very typical of many other HPC centres, close relationships with researchers in the OeRC enable us to trial innovative or new platforms, such as FPGA or Power8, in a timely and real world ‘research’ scenario.
From this, we learn what we should build business cases for to take forward into production services, or where we need to further engagement with vendors.
Is there a skills shortage? Why is it important to get more students involved?
There is a skills shortage in both HPC development and support and coding for software. The future challenges in HPC lie not just in the development of newer and more innovative hardware platforms, but the ability for researchers to exploit those platforms via software that can take full advantage.
The increasing trend towards multi-core platforms and the understanding needed to program those platforms in the best way – to work around some of the limitations of memory bandwidth and interconnect bandwidth limitations to achieve the ever-predicted higher and higher performance levels – needs smart software engineers.
Every year, hundreds of undergraduates pass through UK universities across a broad range of disciplines that at some point utilise computing software to solve something. At postgraduate level this trend continues yet post university when trying to recruit for people interested in further developing or applying computing skills in an HPC or advanced software environment and the lack of applicants for jobs is often staggering. So where do these smart students go?
The increasing demand on HPC from those looking to do some form of data mining/data analytics is pushing newer disciplines towards using HPC. Yet for most of these new researchers they are often oblivious to the benefits that HPC may bring, as they have never had any exposure to it at school or undergraduate levels.
Do you see any differences in the uptake of IT courses between men and women?
It is notable that women are almost missing completely from the HPC landscape. In general, in computing research, we see a higher proportion of women. The ARC facility is physically based in the Oxford e-Research Centre, a multi-disciplinary centre that supports projects across a range of disciplines from Anthropology to Zoology.
Within the centre, there is a higher proportion of women than perhaps seen elsewhere in computing. The research side seems (anecdotally) more interesting to women than the more service side of the IT business.
Overall though, when recruiting there is a decrease in those interested to come and take up research or IT support related posts. Whether this is because of reducing numbers or a healthy economy paying better salaries in the commercial sectors is less clear – but often thought by some to be the case.
Can you tell us about any interesting research projects that the new HPC will be supporting?
The biggest, newest project that the latest HPC service is supporting is the NQIT (Networked Quantum Information Technologies) project.
This is being led by the University of Oxford and aims to be, amongst other things, disruptive to the world of computing. To design, create and overcome some of the challenges that have faced quantum computing to date – and to make possible the creation of a working and useful quantum computer.
This has the potential to impact everything, the potential to change the future and direction of HPC, and, with it, what can be achieved in perhaps helping to solve in the future some of the grandest challenges we face today.
Traditional computers are heading towards bottlenecks that perhaps quantum computing can break to enable a step change in modelling, analytics, etc. in the future. A challenge will be perhaps predicting just how near that future may be.
With IT, and HPC, moving so quickly, have you seen your role change at all – for example to include assisting researchers in different ways thanks to the advances in technology?
Our role has changed over the last ten years from just providing access to hardware platforms for those that knew how to exploit them – and in the process meaning that we were just effectively keeping the lights on – to now being more focused on the researcher themselves and helping them to adopt the best approach for solving their research problem.
That may mean pointing them towards other resources, such as services in the cloud, or advising them how best to use their existing application on a desktop and that actually HPC will bring no advantage at this time for them.
In other areas, particularly humanities, we are increasingly helping users to understand how to start using HPC, to help them actually understand what their research workflow is, and how best to use a particular software application to solve it.
Thus, increasingly, training, support and direct engagement with researchers is becoming far more important to the overall business of the University than just running a HPC hardware service.
Are there any advances in particular that have helped?
At a university level, the simple commoditisation of hardware has made it much easier to build ‘plug and play’ HPC systems. Whilst national scale facilities still require a higher level of expertise to architect and design and maintain the service, at a university level the quality and compatibility of hardware has made it much easier to buy off-the-shelf HPC systems.
They are not perfect – and they still require love and attention beyond what a departmental server or desktop would require – but increasing amounts of staff effort can now be spent on supporting the researcher rather than the hardware.