In February 2013, the Foreign & Commonwealth Office Science and Innovation team, together with Research Councils UK (RCUK), organized a delegation of High Performance Computing (HPC) experts from the UK to visit the USA to (1) identify areas and mechanisms for collaboration between the UK and USA; and (2) identify lessons the UK could learn from the USA HPC community - especially with regard to engagement with industry and the software challenge. This report presents the findings of that delegation, including recommendations for action, along with a summary of the visits and meetings occurring during the week.
UK-USA Consensus on Key Issues
Before describing the findings on collaboration and lessons learnt, it is worth highlighting some common themes emerging from the week, upon which there is strong agreement between the practitioners in the UK and USA:
A. Effective High Performance Computing cannot be delivered by hardware alone – indeed the highest impact benefits to research, industry and the economy will come from HPC related innovations in the software applications.
B. Development of HPC skilled people is required to achieve the optimum economic and research impact from HPC. There are two aspects to this: (i) radically expand the numbers of researchers and businesses aware of HPC; (ii) a substantial program of development of HPC skilled researchers and HPC practitioners. It is widely agreed that the country that gains leadership in these human aspects of HPC will enjoy a significant competitive advantage in the economic benefits of HPC exploitation.
C. Significant economic benefit can be gained from effective engagement between industrial users of HPC and centres of HPC expertise - including academic centres (e.g. Oxford e-Research Centre, OeRC), publicly funded laboratories (e.g. the Hartree Centre), and UK commercial providers of HPC expertise (e.g. NAG Ltd.).
The delegation considered the mechanisms and opportunities to improving collaboration between the UK and USA in the area of HPC, focusing primarily on academic collaborations. The delegation found that:
D. There are many existing UK-USA collaborations (such as the G8 Exascale initiative) and established mechanisms - e.g. with the US National Science Foundation (NSF), the US Department of Energy (DOE), and others.
RECOMMENDATION 1: RCUK should better promote these existing collaborative routes, especially within all relevant call documents.
E. Concrete projects are needed to foster true collaboration, rather than just overall desires to collaborate.
RECOMMENDATION 2: RCUK should find a way to use the established international links of the UK’s key HPC organisations (i.e. Hartree, Edinburgh Parallel Computing Centre, NAG, …) to support computational researchers to develop specific collaborations.
F. This visit was deemed to be successful but only covered the north eastern parts of the USA – there are significant centres of HPC activity elsewhere across the USA.
RECOMMENDATION 3: Follow-up visits (ideally with the same team) should be arranged to other parts of the USA, notably the southern central states, and the west coast.
Lessons from the USA HPC Community
The delegation considered what lessons the UK HPC community could learn from the USA HPC community. The delegation found that:
G. The NSF has operated an industry engagement scheme (not HPC specific) for around 25 years. The Industry-University Collaborative Research Center (I/UCRC) program has several aspects that UK funding agencies and academic centres could learn from.
RECOMMENDATION 4: RCUK should review the NSF I/UCRC program for elements to adopt in the UK.
H. Matching the experience of the UK, the USA HPC community has learnt that getting the best science from HPC requires a greater focus on stable long term funding for skilled HPC practitioners in addition to the hardware systems.
RECOMMENDATION 5: The UK should explore how to establish and expand stable long term funding for HPC expertise in the UK.
I. Effectively engagement between centres of HPC expertise and industry users is beneficial to both parties, but is also difficult to get right. Ensuring the maximum benefit for industry requires a specific focus to the engagement by the HPC expertise provider, which can often represent a culture challenge for academic providers. It is also important for industry to be able to access variety of providers, technologies and expertise.
RECOMMENDATION 6: BIS should undertake an independent review of the full and diverse set of UK providers of HPC expertise/facilities involved in HPC industry engagement and how to best use these (including Hartree, HECToR/ARCHER, the regional e-infrastructure centres, NAG Ltd., etc.).
J. The USA has deployed several novel HPC architectures in anticipation of future technology directions and/or emerging user communities – e.g. the systems at Pittsburgh.
RECOMMENDATION 7: RCUK should review the roles of the various novel HPC facilities (including the Emerald GPU system, those at the Hartree Centre, and others) with a view to ensuring the UK computational research community is best prepared to exploit future technologies.
Gerard Gorman, Imperial College London (Editor)
Mike Ashworth, STFC Daresbury Laboratory
Andrew Jones, Numerical Algorithms Group Ltd. (NAG)
Lee Margetts, University of Manchester
Andrew Richards, University of Oxford