cluster-fsck

[notmatt]

Gya-hah! It was the cluster. I got hold of the admin and he confirmed it was a hardware issue. He rebooted some of the nodes, and I'm up and running again. It's nice when I'm not entirely at fault when things don't work right, but after this I'll have no one to blame but myslef. I'm definitely going to beef up my MPI though. If I'd known what I was doing better, I could have identified this problem sooner and saved a day of computing time.

I still want to get on the Berkeley clusters, since they're newer, faster, and probably more reliable. However, in order for me to gain access, I have to be a Berkeley employee. My colleague there says the easiest way is for him to sponsor me as a Visiting Scholar, despite the fact that I have no concrete plans to actually visit there in person. This appointment will cost $200/year to set up, despite the fact that I will draw no salary from them, and use no physical space. I can only assume it's to pay for the electricity and bandwith I'll be taking up.

Comments

[rbandrews]

And to recoup hardware costs.

Have you thought about Amazon's cluster? They're renting it out now; I know the BoardGameGeek people did some serious crunching on there.

[jameshroberts.livejournal.com]

I'm looking to fulfill my computing needs without having to pay for it. The cost of the Berkeley appointment will come out of my sponsor's grant. The other advantage to the local clusters is I don't have to bother with scheduling. There are few enough people competing for resources that I can just grab the nodes when I need them, and release them back into the wild when I'm done.