Supercomputing 2006 – A vicarious look

I have attended SC 2006 vicariously, mostly through blogs such as scalability.org and clustermonkey. From what I have read there and other places, here are some of the things that jumped out at me.

Accelerators are in. Maybe I have been barking up the wrong tree, cause there seems to be large number of accelerators at the show. An old favorite, MD-GRAPE was there as well. I still think it would make a great system to do accurate binding energy calculations, but I have never run into anyone who ever wanted to buy one. I’ve had the pleasure of playing with a haptic interface hooked up to an MD-GRAPE system. Loads of fun for protein simulations.

Power to the people. In one of Joe’s posts, he talks about “making more power open to wider groups of people”. I cannot agree more. We need to be in a situation where the average person who wants a specific task done should not be hardware limited (within reason). This includes the scientist who wants to visualize complex scientific data as well as the researcher who wants to run massive MD simulations. I don’t think everyone looks at the technical specifications of the hardware these days. People look at form factors, ease of use, etc. Perhaps the user doesn’t care where the computation is happening as long as they can run it and look at the results. All these needs point to two areas that I have actually criticized in the past, accelerators and grid computing. I believe that at this point in time both have serious issues with commercialization and acceptance, or perhaps users aren’t quite pushing their current systems in a way that makes them wish they needed these technologies. I also suspect that there is a level of awareness, or lack thereof, that plays a part in adoption as well.

Clusters still rule. This is obvious from many of the reports out there. Based on recent trends such as Amazon’s web services, discussions with some friends and listening to people like Jonathan Schwartz, I am suddenly not sure what the long term future of the cluster is. I am a huge fan of clusters and they are still, in my book, the best way to do any serious computing in a controlled environment. But the day of the large compute cluster seems to be fading away, with power, space, virtualization, etc becoming the dominant themes.

Odds and ends. A lot of people seemed to be talking about Panasas and their cluster storage products. The latest Top500 list is out. A few changes, but the Blue Gene at Livermore still tops the list. Sun had a strong showing too, just months after I had written them off. Penguin Computing showed off the Sclyde Life Science Suite and there were FPGA type apps from companies like SGI. If anyone has anymore titbits, I would love to hear, especially in the realm of biomolecular simulation and bioinformatics.

Technorati Tags: Supercomputing 2006, High Performance Computing, Scientific Computing