katei.fi is down, or, magic smoke
13 Sep 2010 21:24:37 hardwareMy old new server from two years back let the magic smoke escape just
earlier today. :-( chikan.katei.fi
was churning in hundreds of thousands of
lines into PostgreSQL database when disaster struck. When the process stopped,
at first I thought it was yet another hickup with chikan’s integrated network
card, but when my automagic script didn’t kick in and save the day, I got
nervous.
After some checking I found out that chikan had powered down – uncontrollably as it later turned out. When I approched chikan I smelled unmistakably smell of magic smoke. I unchained the poor thing and carried it to the operation room. I disconnected all non-critical components and switched on the PSU. The only reaction I got was a frantically blinking green LED on the motherboard. I soon came to realise that this was something that could not be solved just like that.
At first I digged out tomodachi, my old server last powered up in 2004, but after booting it up came to conclusion (thanks, zeska!) that it would be easier to throw the disks to dante. With a 2.2 GHz AMD64 chip it could be considered an upgrade over chikan’s passively cooled Celeron 220. After some tweaking I managed to get dante to accept chikan’s disks and boot up. Without getting into details, after two hours dante was ready to take over chikan.
Switch to dante is a two-edged sword. On the other hand, dante is significantly faster than chikan, processing database queries at least four times as fast. However, on the other hand, it does have moving parts, including fairly noisy chassis fan. Besides that, chikan was much more power efficient. Right now the question is what happens next. dante can deal with katei.fi and intra.katei.fi just fine long into foreseeable future, but somehow I don’t like the idea too much.
What kind of hardware should I get next? If the three factors to consider are reliability, price and performance, I opt in for the first two. I thought I had this covered already with chikan, but obviously I was wrong. I’d love to hear your ideas on this.