IT Consolidation and Incubation – Watchwords for HPC

A conversation with Satoshi Matsuoka, Tokyo Institute of Technology.

This summer, the new TSUBAME Grid Cluster at the Global Scientific Information and Computing Center at Tokyo Tech rocketed to 7th place in the TOP500 List of the world’s fastest supercomputers. Built on Sun Fire x64 server technology with 10,480 AMD Opteron processor cores, the computer sustained 38.18 trillion floating point operations per second (TeraFLOPS). CT talked with Tokyo Tech professor Satoshi Matsuoka about high performance computing infrastructure for higher education environments:

Is the TSUBAME supercomputer at Tokyo Tech now the fastest university-owned computer in the world?

It’s the fastest university-owned computer, and the fastest civilian computer open to the public.

Satoshi Matsuoka

Satoshi Matsuoka at last month’s
Sun Microsystems
server intro in San Francisco

Given that status, perhaps you could offer some advice. What is important for other universities to consider, regarding high performance computing?

Consolidation with the IT infrastructure. It’s critically important for two reasons:

One is that we need to be able to incubate users to do high-end science. So, we have a machine like the one at Tokyo Tech, demonstrating that at an institution like ours, which is not big (we only have about 10,000 students), we can, within our budget, excel in performance. Then we need to provide this power not to just 10 or 20 people, but to everyone in order for them to be incubated towards being high-end users.

The second is that many institutions, despite the fact that they have lots of good faculty members in science and engineering and in other disciplines who could make use of high-end computing platforms, have facilities that are quite fragmented.

And while some institutions may have attempted to start consolidated IT infrastructures just by doing more administrative types of IT activities [on high-end machines] – like e-mail, or Web serving, or student ID management systems, or ledger – that’s a waste, because in those areas you don’t need much hardware though you need lots of people. Instead, if you have [a high performance system] that is scalable, then for a very nominal cost you can increase your infrastructure – to be centralized, simply managed, and also fairly major in terms of compute power.

So then with such a scalable system, you would fold the administrative computing in with scientific computing as a part of your plan for a consolidated IT infrastructure.

Exactly. Now that’s not to keep people from having ownership of machines, but the point is that IT consolidation has been largely ignored as a computing area. You could in fact think of computing as just one service. By consolidating IT we can see lots of benefits. And even for institutions that really can’t do that themselves, we can outsource and provide services in a consolidated fashion.

Satoshi Matsuoka is a professor at the Tokyo Institute of Technology, Global Scientific Information and Computing Center.

comments powered by Disqus