dcsimg

HPC 2010: The Good, The Bad, And The Ugly

A look at some disappointments and highlights of 2010 HPC season

Now that the year has rolled over, I find myself wanting to sit back and offer some kind of 2010 recap. Lately, it seems there are a fair share of failure articles about the consumer technology market. These articles often have titles like “The 10 Most Stupid Ideas of 2010″ or “Biggest Tech Failures of 2010.” I consider these traffic accident articles because most people tend to like bad news more than good news. When was the last time you saw traffic slow down and look at someone who was being nice.

Having thought about it, I really don’t want to write an article shaming people in the HPC sector. At the same time, I must say there were some disappointments for me this year. Therefore, I’ll mention some of my personal HPC highs and lows in 2010. I’ll also drop in links to my previous coverage of these events should you like a little more depth.

Let’s start with SC10. A good event and by all indications the industry has weathered the economic down-turn and is on it’s way to better days. This feeling was confirmed at the IDC breakfast. (The fact that IDC even had a breakfast was an indication things were getting better.) There was a record for technical program participation (4,390 attendees) while overall attendance broke the 10,000 mark. For me there were two disappointments at SC10. I could not find any place that was showing the Pacquiao-Margarito fight on Saturday night (November 13). I checked Harrah’s Casino and every other sports bar I could find. No luck. I’m a fan of the Pac-Man. The second is that I was reminded that I don’t have any good New Orleans restaurants my hometown.

The SC10 Beowulf Bash was a also a highlight. We had over 750 people show up! As has become our yearly lament, “we need a bigger venue next year.” Score one for the community. Another, highlight that I just stumbled upon is the SC10 Analyst Crossfire session. This is a video well worth watching. The panel includes Addison Snell, CEO, Intersect360 Research (moderator), Michael Wolfe, Engineer, The Portland Group, Inc., Peter ffoulkes, VP of Marketing, Adaptive Computing, Jay Boisseau, Director, Texas Advanced Computing Center, and Thomas Sterling, Professor, Louisiana State University.

For me the biggest ugly of the year was the dumping of SUN’s HPC division by Oracle. Of course, they bought SUN and can do what they want, but just because you buy a house, does not mean you bulldoze the parts you don’t like. A good community perspective comes from Joe Landman (read comments as well). Closer to home, I have concerns about what is now Oracle Grid Engine and the fate of the open version. The SUN Open Source license is some insurance, but again, why trash something that is working. The same goes for Lustre. Sigh.

Some of the best news this year was in the form of announcements. I am eagerly waiting the first AMD Fusion processors (video logic on the CPU die). AMD has been working on this for the last few years and Intel has announced the new Sandy Bridge processors will have embedded video as well. As I understand, they both will have OpenCL support. I covered this trend recently because I assume all good things eventually get absorbed into the CPU die.

The other exciting announcement was the Intel 48-core processor. I talked about why this should make MPI programmers happy. This chip could be significant. test bed for the future as it addresses “hitting the core wall.” In other Intel news, I found the Hold On A Minute paper comparing optimized multi-core and GPU performance reassuring. I don’t have a favorite in the GPU vs Multi-core argument, just a penchant for good numbers and reality. You have to buy the paper, but you can read my take on it for a summary.

Perhaps my biggest disappointment was the lack of progress on the things that are really needed to move HPC forward. These include, education, better software tools, easier management, and more people. The same things that were identified in 2004 in an IDC Survey sponsored by the Council on Competitiveness (COC). In a sense these issues were “formally identified” by the COC study, the situation was and still is common knowledge in the HPC community.

From a technology standpoint, we seem rather complacent. We are cramming more cores into the same space and at the same time adding GPUs cards in a somewhat orthogonal fashion. While GPUs as array processors may sound new the concept is quite old (and works in many cases). We are able to push the limits of computation with the Top500 list while at the same time we cannot provide the “missing middle” with what it needs to grow. I also do not believe this is an economic argument i.e. that there is no market for HPC with small and medium size organizations. The opposite is true. There is a crying need for HPC in industry because innovation allows companies to compete. And all companies need to compete.

I am somewhat excited about Cloud HPC. It does make it easy to get cycles on demand, but I think it is not take off until we address some of the issues mentioned above (and for those that promote it calm down). By the way, I consider Cloud HPC to be dynamic provisioning as opposed to virtualization. HPC needs to be close to metal.

My biggest excitement this year was around the progress in low power processors. I think the low power market will offer some new DIY options for HPC. Cell phones and tablets may be a hardware driver like web servers were for today’s clusters. I am not advocating cell phone or tablet clusters, but the use of the low power goodies that are inside. Remember, the first HPC clusters were about not following the rules. Let’s keep that in mind as we move ahead into the decade.

Fatal error: Call to undefined function aa_author_bios() in /opt/apache/dms/b2b/linux-mag.com/site/www/htdocs/wp-content/themes/linuxmag/single.php on line 62