In this video, Moderator Bob Feldman hosts a session entitled: Supercomputing: Where to from Here? Recorded at the National HPCC Conference 2011 in Newport.
Dr. Eng Lim Goh, SGI
Bill Feiereisen, Intel
Shumel Shottan, BlueARC
Steve Lyness, Appro International, Inc.
Marc Hamilton, HP Americas
Below is a summary of what is told. It is just my notes, so go to the times mentioned to listen to the exact answers. Some details I did not write down, you might think are important, but I did not (or missed as I English is not my mother-tongue).
Q1 (7:30): we have achieved a lot, what’s next? The most important things that will influence the first 5-10 years.
A (8:50): Mostly big data and software will tell (software first), not hardware. Power&cooling. Limit is bandwidth. Smart business-models.
Q2a (23:40): Is a pursuit of exascale wise for the USA?
A (24:40): Other countries invest too. Government-funded and corporate funded. It is good for the economy, so it goes as it goes. National effort
Q2b (31:15): Will exascale contribute to sub-exascale HPC?
A (32:00): Yes, will flow down.
Q2c (36:10): should exascale be prioritised? Self-answer after watching faces of the panel: Ok, I don’t think so.
PQ (37:00): Software, software, software. Is exascale defined by hardware?
A: yes, software pushes it.
PQ (38:20): Is the focus on exascale too much?
A: focus is on continuum, second on exascale. You should have a goal. Also: scientist don’t go on the one-line from petascale to exascale, but goals and methods diverge. “Exascale is without the label”, so 6th generation HPC.
Q3a (42:40): do we still need the highly trained engineers to help scientists the coming years, or can scientist get their hands on the tools to operate the HPC-systes themselves?
A (34:30): Both. And also no: the latest technologies will not have the tools ready for to have the max speed. Also: it’s a choice ike we did with cars; we can choose for hiding complexity or go for the extreme. We need to find what a higher level form of MPI is. Extending the analogy: We have both cars and F1.
PQ (54:30): Computerated MPI? A: Yes, would be nice. All agree it is an interesting approach. LISP? Library-based, so scientist can choose, drag and drop. It takes time… (HP-guy quotes results from a MS-research which are known to be tweaked to favour Windows HPC; bad Mr.Hamilton!)
PQ (1:01:50): What would be the HPC-applications for the normal guy? A: in 20 years what we have now on HPC. We will not see it coming. Personal HPC, like personalised weather report and personalised medicine.
1:06:45 PQ: Should the government investment in exascale be more justified? A: No. There are still problems which can be solved with HPC: weather, earthquake-prediction, safer cars, etc. But PR of HPC is not that good the latest years.
Q4 (1:11:40): What about the Chinese? What about USA vs China, or China in cooperation with the VS.
A (1:12:40): Should we be surprised? USA now more Art-history-majors. “What is that Chinese super-computing doing today?”. We should focus on productivity per FLOP now. There is competition and… there is growing competition.
Q5 (1:18:10): Should we focus on keeping and controlling IP?
A (1:19:55): Focus on open standards, because we (US companies) sell in China. There is open, government IP and corporate IP. It is more a competition between companies instead of competition between nations.
Q6 (1:23:15): Because of education, HPC-companies are all over the world. Will we see fewer HPC-jobs in the USA?
A (1:24:50): Not worried about the jobs, because there are enough jobs. Should address the problem that students leave the country, once graduated. Problem is more how we motivate students to become HPC-scientists.
PQ (1:31:40): Dont we have already P2P-Exascale combining all smartphones and laptops in the US?
A: Ok, thanks! So problem solved.