One can survive on just ten
Hours of sleep a week?
Well, I wouldn't've. But it turns out that, not only is it possible, it's a helluva lot of fun! The Cluster Challenge this year has been a really different experience from last year, but these German fellows are a (metric) shitton of fun. :) We've spent a lot of time with the cluster, but it's also been a good time to, like, learn about other cultures, yadda, yadda. But, seriously, it's been interesting. They've not missed a chance to make fun of me for the US university system; I'm missing a test and a couple homeworks this week, and my professors requested that I do them and fax them, which is apparently completely a completely unreasonable idea in their system. I guess I thought it was sort of unreasonable, too. Grr. Anyway, we have been having a great time. The organizers set us up with a trip to the Main Event, which is a place near Austin with laser tag, rock wall climbing, bowling, an arcade, pool tables, ...
The challenge itself went well. We ran more (I think?) data sets than any other team, and we didn't run into any issues. The MIT team didn't show up, and the StonyBrook team from last year essentially bailed them out, with help from Dell. The MIT rack didn't show up; they thought that they were going to run a rack with GPUs in it, but there was no chance that they'd finish in time to present it here. So the SB kids who were supposed to just be volunteers asked Dell for a rack and some nodes since they got here. They scored an impressive... 16 GFLOPs on Linpack. Ouch. Arizona is running a Windows cluster, which is an interesting crippling in itself. The Colorado team was running some hardware that was scheduled to be announced, but it didn't wind up happening, so they had to rebuild their cluster the morning the competition started lest they violate their NDAs. Awful. But the other teams (Purdue, Alberta, and Taiwan) seemed to do alright.
There's about a thousand other things that I could say, but I have to go and not sleep some more. :)