SC07 Cluster Challenge - race day

by Brent Gorda


After almost a year of preparation by the committee, a week of on-site infrastructure setup, and 2 days of on-site prep for the teams: we are finally at race day!

Tonight at about 8pm at the SC07 conference in Reno, we will officially kick-off the first attempt to build and race cluster-based supercomputers on the conference floor. The room is huge (20,000 ft^2) and beautiful, being decorated with banners, posters and even an airplane courtesy of the Reno Air Race Association.

If you are at SC07 and in the technical program, we invite you to visit us in hall 1 @ 8:00pm for the start and to cheer on your favorite team.

Brent Gorda


Cam Macdonell
2007-11-12 09:35:24
Can you smell the silicon burning?

The nodes are hot and the team is excited. Staying under 26 amps is a real challenge and our team is doing are best to walk on the edge. All the teams have been working away since Saturday getting the gear running - it's definitely a high voltage in many ways. Although Indiana U's toy racecar track definitely reflects how much fun everyone's having.

Even though the competition is less than 48 hours, it sure feels a whole lot longer! I can't wait 'til 8pm.

Cam Macdonell
University of Alberta

Paul Lu
2007-11-12 10:32:30
Good day. Obviously, many of the students are too busy to be blogging. So, coaches like myself will act as their stenographers.

The Cluster Challenge is about many things, including educating a generation of HPC and cluster experts.

Towards that, one of our students made the comment that a noteworthy revelation of the whole experience is that the power draw of a cluster can vary greatly (e.g., single digit amps to double-digit amps) as the specific workload changes. Even the different HPCC components can have significant differences in power draw.

Of course, many HPC and sys admin experts know this, but they likely didn't learn it as part of the typical CS undergrad curriculum. More than likely, this is still something learned from hands-on experience, thus the need for the Cluster Challenge.

And if some expert wants to write a good paper or tutorial about this very topic, I would like to read it myself!

Cheers, Paul Lu, University of Alberta.

Preston Smith
2007-11-12 14:29:39
Team Purdue's enjoying all of the activity, and being around all of their peers and getting exposed to the best and brightest in HPC..

Our students have gotten into the spirit of the occasion, even trekking down the street to a nearby hardware store to procure a Boilermaker-esque sledgehammer to make our space stand out.

Preston Smith
Purdue University

Check out our blog of the activities at

Bob Beck
2007-11-12 14:40:30

And as the excitement builds we just had a mad rush from all the
teams to "redeploy" some of the sofas and chairs from the lounge area
to their booths.
Doug Smith
2007-11-13 08:02:14

Kudos to Brian and John of team CU Stampede for pulling the first night shift of the competition. After submitting the HPCC runs we have completed several of the data sets ahead of schedule and our cluster purring right along.

A pat on the back goes to team leader Dustin for optimizing HPCC and getting our team off to a great start. Our HPCC benchmarks were submitted ahead of schedule.

Things I heard this morning from the night shift:

"We stayed under 26 amps all night!"

"I'm glad theirs not a mountain dew limit on the competition!"

Preston Smith
2007-11-13 08:32:42
The Purdue team's moving along, completing HPCC relatively quickly, achieving 83% of peak.

Now the HP/AMD cluster is doing all of the hard work, and when I arrived this AM, I see students exhaustedly sleeping on the "liberated" sofas mentioned in somebody's earlier post. It doesn't appear that the coffee pot in the back has brewed anything yet, but I suspect that'll change as more of the team members begin to stir!

Preston Smith
Purdue University

Bob Beck
2007-11-13 18:40:25

As many of you may have heard, supercomputing 07 experienced a site wide
power outage at the reno convention centre today. This of course included the
cluster challenge where all teams lost power on their running systems.

As someone who professionally deals with operational issues in a large data centre I'd like to congratulate not only the University of Alberta team, but all the teams on their job of recovering machines and restarting under time pressure. Watching from the "other side" of the wall is something
I don't get to frequently do in that situation, and seeing groups of students handling it cooly and professionally was very nice to see.

Congrats to all of you.

2008-06-15 01:27:06
2008-06-15 01:29:07
2008-06-15 01:31:26