Ben Kudria » Weblog

Whew, what a day.  It’s 9PM here in Reno, and I’ve got the first shift on the floor.  One hour ago, the challenge overseers invited all the reporters and conference attendees into the exhibit hall, projected a large countdown, and waved a flag, all to culminate in…our team captain pressing a return key.  My roles for the challenge are POVRay (more below) and visualization, so our large TV immediately springs into life, green progress bars jumping and flashing.  Screenshots are below.

Update:  10:30 PM:

So, a bit about our configuration: we are running a 13-node cluster, with each node containing 2 quad core Intel Xeon’s, each core clocked at 1.86 GHz.  Each node is connected to the network with both a gigabit Ethernet connection, and a 5Gbps Infiniband link.  There is one hard disk located on the head node, for use by the entire cluster.  Data from the applications is first written to a 8GB RAMDisk shared over NFS.  Our cluster has a theoretical computing power of 750 Gigaflops, or 750 billion floating point operations per second.  For the contest, we are limited to 26 Amps total for the power, split over two 13-Amp circuits.  Power is distributed to 2 monitoring PDUs (Power Distribution Units), and if we go over the 13-Amp limit, alarms go off and we get emails.  We currently idle at around 8.1 Amps, and we are at 12.5 at full load.  I’ll be able to post some more pictures of the configuration, and the conference floor, after I get back to the hotel tomorrow morning.

Update: 11 PM: 

The contest is structured in two parts: a benchmark and a application and data set computing challenge.  We are just now finishing up with the HPCC benchmark suite, and after that, I’ll begin the POVRay jobs, rendering either pollen particles, SQUIDAtom (not sure, it’s the last image, below), or any other strange number of things.  Our two other applications are POP, to simulate ocean temperatures and curents, and GAMESS, which does crazy molecule and atomic simulations.  We have 25 separate datasets over these 3 applications, and each data set is worth some number of points.  Points are also awarded for the HPCC benchmark.  At the end, the teams with the most points wins!

Update: 11:45 PM:

I’ve finished rendering Pollen6, a nifty simulation of what looks like the most painful pollen particle I’ve ever seen.  I’m rendering the SQUID data set now.  After that, I’ll try what is labelled Scene005, which I believe is a rendering of microbial activity.  The next shift comes in at 6AM,  at which point I’ll grab a cab back to the hotel, although there are some inviting leather couches nearby.

So, the cruise kinda sucked.  First, our bus broke down, then it started raining, and finally, the boat (The Mason Dixie II) wasn’t that impressive.  The food was good, but we had to endure "entertainment" by a fake Sam Clemens a.k.a Mark Twain.  The jokes were not funny.

On the plus side, we got our cluster up and running, and I was even able to run POVRay yesterday.  So far, we don’t have any problems fitting under the power limit.

We also get a couple of perks: since the conference organizers weren’t able to secure prepaid Visa cards for us, we get an envelope of cash every morning.  There is a party for all the Cluster Challenge participants Wednesday night, featuring a performance by The Blue Man Group.  We get extra points because we have Intel processors.

We also have two very tall Stony Brook Banners, a table for our TV, small stuffed mascots, and other schwag.  Our booth space is looking professional.  The other teams seem to be doing well.  Most have half-height racks instead of full-height, and the IU + Apple team has very nice-looking equipment.  I maintain that we have a good chance of coming out ahead.  Watch this space.

So I’m typing this from the back room in the Reno-Sparks Convention Center in Nevada. We’re almost done setting up our full-height rack with all our computing nodes. We’re hooking up the 5 Gbps Infiniband as I type. After we’ve got the hardware all set up, we still have a bunch to do on the software side of things.

The other teams are arriving now also. The Taiwanese team is already here, they are just waiting for their rack. The Purdue team has shipping containers strewn about, but their team isn’t here yet. Same with the Indiana University team - their Apple hardware is just sitting there, some eight or 10 XServes, plus a paltry 23-inch Cinema Display. Our Plasma TV is 42 inches, and it’ll be displaying all our progress, as well as the slickest screensavers I can find.

After that, we have a all-hands meeting to go after safety procedures, and tonight, when everything is ready, all the teams are going on a chilly late-evening dinner cruise on Lake Tahoe. It sounds pretty fun, I’ll tell you about in my next post.

In a couple of hours, I’ll be waking up to catch a flight to Reno, by way of L.A. That’s right folks, behold the efficiency of the American Airlines flight scheduler. Anyway, in Reno, my teammates and I will…wait, what? Teammates?

Right, so in the past month or so, a groups of us undergrads here at Stony Brook University have been building a kick-ass supercomputing cluster. That’s right, tens of thousands dollars worth of equipment, with Debian, NFS, Infiniband, Xeons, and more, to create a truly powerful computing cluster, capable of rendering pollen particles, simulating molecules and oceans, and defeating benchmarks in a single bound. I’ve been busy helping make it all happen.

We’ve been preparing for the SC07 Cluster Challenge, an event to be held next week in Reno, on the SC07 conference floor. Six teams of eager undergrads will be competing to run the benchmarks faster, and to impress the judges more. Wish Stony Brook luck!