Murphy was, of course, right

[fivemack]

For the last several months, I've offered some disc space and some compute cycles on one of my computers to pseudomonas for developing thesciencebehindit.net, a site which uses some very fancy natural language processing to attempt to work out from which scientific paper an article in the popular press was written.

Today, the site was mentioned on badscience.net.

This evening, I noticed that the computer had dropped off the Internet, and that the air in the study smelt of that magic smoke that electronic components emit as they fail. I happened to have a spare power supply, and with this replaced, and the fuse in the power supply changed, the computer started working again.

The dead component was a Colorsit EN60950 Switching Power Supply; I think it's quite ancient, it may well have come out of a computer bought at the office in 2004 to be plugged into a case that I bought in 2007. Maybe five years is a reasonable innings for a power supply ... the last one that I had fail did so after six months.

To the pub now, I think, leaving for tomorrow the question of why compute node #2, which yesterday was happily transferring tens of gigabytes around by NFS, today denies having an ethernet interface.

Comments

[(Anonymous)]

That's quite cool. The entity extraction seems to work well - pity it only knows about biological things are present though. I wonder what luck one would have using these entities in a Google Scholar search. Ed.

[pseudomonas.livejournal.com]

Pretty good, but unfortunately I think Google's T&C preclude doing that on an automated basis.