(resending my reply...)

On Fri, 1 Apr 2005, IOhannes m zmoelnig wrote:

Tim Blechmann wrote:

...

hi gem-devs ... i experienced something very strange with gem on a few machines ... i'm using gem to display images and play back videos ... but on one point (after a few hours), it refuses to load images / videos ... the strange thing ... i was able to load exactly these files before ... the version of gem is 0.90... any ideas?

the last time i had this behaviour, it was the first sign of an upcoming hard crash of my harddrive. probably you better run...

Hi. I have a different warning for you, in a similar situation. Let me explain.

On the last installation I worked on (a museum in Quebec City) I've had a problem with one of the subtitles, which was encoded in QuickTime-PNG (because I needed alpha), and then the PNG decoder would yell "CRC Error!" and then crash because I didn't have an error handler for *that*. Well it's a good thing it crashed, else I wouldn't have noticed and it would've been very bad. But then it wasn't crashing an hour earlier, so what was going on? I rebooted and then everything was fine again.

Turns out that on a 512M RAM bar there was *1* bit which was occasionally failing, and the disk cache chose that spot to hold a compressed frame of the subtitle. It also explains a corrupted directory (that fsck wouldn't handle!) that appeared upon installing Linux, and some files being mysteriously different on backups than on the originals.

I strongly recommend testing the RAM with http://memtest86.com/ or similar. However I also fried a CPU last summer, although it only got up to 80 celcius (AMD specs say 95 celcius max). The difference is that CPU failure will look like random noise in memtest86 (change every time, no pattern in addresses), whereas RAM failure will show definite patterns (the errors stay the same and might even show periodicity, for example exactly 256k apart).

On a related note, I just had a major hard disk crash (accompanied with feelings of guilt about backups being too old). However, the disk had suddenly died, so it was only the controller. Surface damage is more insidious but also you have more time to do backups. I would like to thank Christian Klippel (the multio/vdsp/karma guy) for suggesting swapping controllers with the one of an identical harddisk I happened to have.

.-----------------------------------------------------------------. / The Diagram is the Program (TM) ,--------------------------------} {----------------------------------" http://artengine.ca/matju / `-----------------------------------------------------------------'