HPC has been in trouble with clusters and high end products, according to the speaker. The systems were about 2 billion dollars, but now it is 10 billion dollars, so strong results are being achieved. In 2009 the rise was 35% in the worst year of the recession, Steve Conway showed the audience. If we look at the market by region, then Europe is starting to regain some of the share but it isstill behind.
Servers are bringing in 18 billion dollars globally. As for the forecasts, Steve Conway thinks that the barriers may be formed by the price.
The IDC study results show that financing is available for a software infrastructure for highly parallelized codes.
The development of a business model is needed however. We are to set the framework for creating a European software infrastructure for research and science and analyze the strenghts and weaknesses, Steve Conway lectured. The speaker has done lots of interviews. Library research is being performed by
IESP and EESI.
IDC was very impressed with the European Exascale Initiative because it includes industry. This is not the case with the international IESP, noted Steve Conway.
Some key findings of the IDC study include:
The industrial users have to be involved from the beginning for the codes.
The good news in Europe is that terrific codes have been developed but they are only used by one or a small number of institutions. The dissemination has to be wider. Hardware is easy but software is hard.
Software has to be robust to last for 20 or 30 years.
The overwhelming majority said that they want to work within a EU framework, according to Steve Conway. Expertise is important.
IDC has formulated some recommandations:
1. put a strong focus on the software for exascale development
2. the innovation gap has to do with R&D altogether
3. the HPC spending gap is a lot more severe in Europe than in the rest of the world
4. development has to be recognized by political leaders
5. funding agencies focus on weather and climate research; clean and sustainable energy; automotive and aerospace design; bio-life sciences; particle physics; and materials sciences
HPC leadership is not that important in hardware but in software, it definitely is.
As for the governance, a body should be created within the EC to co-ordinate the HPC strategy in hardware and software to pull together the threads, noted Steve Conway.
The domain expertise resides at the HPC sides. IDC recommands centres of excellence for the targeted domains for planning and funding from the top down. For example: the automotive domain should be split in subdomains.
There must be some way of disseminating the software. The model doesn't work. You need a model for 10 to 15 years progress-dependant and a road to commercialization, broadly defined.
You need a kind of a store fund where people can put up their codes and others can get access to, also to hardware and expertise. The venture capitalists should come to this store fund.
Centres of excellence should not work in isolation. Tigre teams should be set up, very small teams to improve HPC software across Europe by paid assignments. There should be 100 of these teams formed in a couple of years, as the speaker pointed out.
There has to be a split-out of who should pay for what. The EU member states work with a 50/50 funding scheme. The centres of excellence would be about half of that funding amount.
The industrial model as opposed to the scientific model is a requirement because there are a lot of differences, Steve Conway concluded.
The study, "Financing a Software Infrastructure for Highly Parallelised Codes", was conducted under the SMART 2010/0052, contract number 30-CE-0394162/00-51, for the European Commission Information Society and Media Directorate-General, Emerging Technologies and Infrastructures.