HPC User Group (HUG)

Welcome to the NYCA-HUG

New York City Area HPC Users Group

UPDATE July UPDATE

July 17th meeting is canceled.

Please Note: The mailing list has been down and I have been away (vacation). If you need to
contact me directly: deadline (you know what to put here) linux-mag.com I am planing an August meeting 21st meeting. More news to follow.

Our Next Meeting will be August 21 (Thursday) from 7-9PM at Tumulty’s Pub in New Brunswick, NJ (Google Map) The Pub is also located near the New Brunswick train station. (directions) Look for a NYCA-HUG sign or bunch of cluster geek types.

August topic to be determined.

About NYCA-HUG: The NYCA-HUG (New York City Area HPC Users Group) is a group of computer enthusiasts with a common interest in High Performance Computing (HPC) Clusters. We hold a meeting the first Thursday (and some times the third Thursday) of every month where we discuss topics of interest to group members.

To find out more and to join the discussion of other community members please join the NYCA-HUG Mailing List. The mailing list is intended to support the discussions started at these meetings (and elsewhere). You can contact Douglas Eadline directly by using ‘deadline (you know what to put here) linux-mag.com’.

PAST MEETING NOTES

January 3, 2008 Meeting: We had five people attend on a very cold day in NYC. Another good discussion, parallel file systems, benchmarking, and the proverbial question “what is a supercomputer” came up again. Everyone said they would be back again. We had both industry and academic representation. I gave out a draft of my MPI on Harpertown white paper.

February 7, 2008 Meeting: A general get-to-know everyone type discussion, then down to business focusing on file systems, with Doug discussing FUSE (File System in User Space). The Lustre file system came up in the discussion. Interesting tie in, does anyone have further information on Lustre using FUSE? (Thanks to Robert Rusinko of
GridPlexus for the summary).

March 6, 2008 Meeting: We had Mario Juric from the School of Natural Sciences, Institute for Advanced Study at Princeton talk about GPGPUs (General Purpose Graphical Processing Units) as co-processors for number crunching applications. NVIDA has recently released CUDA, an API for programming their GPUs. Mario talked about using GPGPUs and gave a rule of thumb based on his experience. “If your application can use the GPU Functionality (think fast array processing) then you can expect a 10 fold increase in speed with about a weeks investment in source code modification”. In addition, Mario pointed out that current GPGPUs provide 32 bit precision. The 64 bit versions are expected very soon.

You can find more information on AstroPhysics from the AstroGPU site (videos available). You may also want to take at GPGPU and Pixels to PetaFLOPS: How GPUs are Pushing the HPC Envelope as well.

April 3, 2008 Meeting: We had planned to talk about personal clusters, but due to some issues with both the mailing list and some changes on the Linux Magazine website the announcement was late. For those that were there (Robert Rusinko and myself) we managed to solve the worlds problems, but did not bother to write anything down.

May 1, 2008: we had a live hardware demonstration! SiCortex brought one of their seventy two core/processor Catapult desktop systems to the Figaro Cafe in Greenwich Village. The Catapult, is a true Linux parallel/cluster machine that provides the user with 72 processors with a power draw of only 250 Watts. (It can be plugged into a standard electrical service, like those found in the back of a Cafe in Greenwich Village). While the Catapult is not expected to replace the need for larger parallel computers, it is designed to allow programmers to develop scalable codes on a balanced parallel Linux machine. Codes that scale on the Catapult, will run on all other SiCortex machines as well (only faster!). Two representatives from SiCortex were in attendance to answer questions.

June 19, 2008:
Scheduler Smackdown. Torque or Sun Grid Engine You decide. We will be discussing the pros/cons of each scheduler for HPC clusters. Come and add your experiences, wants, and rants.