SMMscale: The Next HPC Challenge

For those that checked in on the state of my thumb, thanks for your concern. If you missed my last column, I gave my thumb a good smashing just before SC10. I managed to navigate the show and New Orleans without any further thumb related issues. The show was a big success breaking last years attendance record of 11,000 people. As always, I have plenty to write about and will be doling out my reports over the next month. And, if you don’t read about it the week after SC, it does not mean the news is less important. There are only so many words, as it were.

I want to address a few issues before I begin this weeks topic. First, I am constantly asked, particularly at a trade shows, how I choose topics that I write about. As with any journalist at SC, I get asked to write about this or write about that all the time. My stock answer is, “I write about what interests me.” I have no desire to try and capture all the news from the show. There are other venues that do this much better than I can. I also like to provide a little analysis and/or opinion about things I cover. In one sense, my coverage is somewhat random. Technology news comes to me in many ways. At trade show it depends who I run into and where I happen to be walking that day. I do like to visit people I know who are doing interesting things, but often times finding something of interest is part of the adventure. Of course, there are always some dependable conference events to attend, which is a nice segue into my next paragraph.

The SC10 Beowulf Bash was another record breaking success. Despite the rain, we had over 750 people attend! (and the boat did not sink) My biggest problem is there were too many people with which I wanted to touch base. Thanks again to our sponsors who made it all happen: Penguin Computing, AMD, Adaptive Computing, Aeon Computing, ClusterMonkey.net, Kove (previously Econnectix), insideHPC, Intersect360 Research, Numascale, QLogic, SICORP, SuperMicro, Terascala, Versant, and Xand Marketing.

As I mentioned I will be covering various topics over the next several weeks, and I want to lead off with something I think as important as the Exascale efforts taking shape (Exascale is 10^18 FLOPS). Unfortunately, it not quite as y as mainstream HPC and does not predict weather, find oil, fold proteins, or collide galaxies. I would like to talk about what I call SMMscale HPC, that is Small and Medium Manufacture Scale HPC.

The success of HPC in big business is . Indeed, according to Jack Dongarra, “Fifty-six percent of the fastest computers in the world are being used in an industrial setting.” There is no doubt that HPC has become an indispensable part of the strategic plan for many large corporations. What about the Small to Mid-size Manufactures (SMM)? You know the ones that study after study say create most of jobs in the US.

I am glad you asked, because there are efforts to address this sector as well. I consider this market sector both a strategic necessity and a lucrative market for all vendors. I assume any HPC vendor would like access to the over 300,000 SMM that are part of our economy. Before, I dig into this idea, I want to point you to Revitalizing Manufacturing: Transforming The Way America Builds. The freely available report is full of interesting information and explains the strategy to address what has been called “the missing middle.” The report is sponsored by NCMS (National Center For Manufacturing Sciences), Intersect360 Research, Microsoft, and Intel.

There are two parts to the report; a strategy for creating a National Innovation Network and some market research by Intersect360. I want to mention the market research first. The use of digital manufacturing can be broken in to four categories; Modeling/Simulation HPC, Modeling/Simulation Desktop, 3D Tools, and 2D Tools. Not surprisingly, 56% of those companies surveyed with 10,000 employees or more used HPC Simulation and Modeling. Of those with less than 100 employees 13.7% use HPC Simulation and Modeling. The use of less sophisticated (and less expensive) 2D tools among this group was reported as 85.4%. There is a clear need for HPC in at the low end. (See the report for more data).

In the first part or the report, NCMS described a plan to develop network of Predictive Innovation Centers (PIC) which would focus on the small and mid-sized manufacturer. The centers will offer tools, training and expertise to facilitate the use of HPC at the “missing middle.” The goal will be to leverage much of what already drives the HPC market. This is a welcome effort and one which I will be watching.

In closing, some of you may be wondering why I’m discussing a rather low-tech HPC topic right after SC10 — where all the latest and greatest HPC wares and news is introduced. That is fair question. As I was walking around the show floor, I marveled at how much really sophisticated HPC is taken for granted these days. In years past, I would have stopped in my tracks looking at a protein rotating in 3D or a global weather simulation. These things are quite common today. Then I thought about how HPC could be used to improve many other areas. To satisfy my curiosity, I ventured over the the Ohio Supercomputing Center booth to check in on the Blue Collar Computing effort. There I found lots of information about the “missing middle” in HPC.

I decided to write about SMMscale computing because I think it is interesting and often lost in the grand scheme of grand challenges that HPC is destined to conquer. In my view, the biggest challenge facing HPC is not Exascale or even Petascale deployments, but rather how to expand and enable HPC to benefit everyone else.