![]() |
a very strange crash: mfaktc cleared worktodo.txt but results are missing
Something very strange happened to one of my mfaktc runs.
You may recall in [url=https://mersenneforum.org/showthread.php?t=24808]another thread[/url] that I was unable to run mfaktc on all eight GPUs in a server without tripping the circuit breaker. Because I don't want to drop these assignments, I decided to finish them using four GPUs at a time. Just for the record: 1) I'm using [url=https://mersenneforum.org/showpost.php?p=526670&postcount=3203]my own script[/url] to launch mfaktc on multiple GPUs and 2) I like round numbers and usually run batches of 100 assignments per device. The first run completed without problems. However, all four instances somehow terminated early during the second run. Both [C]results.txt[/C] and [C]worktodo.txt[/C] show the same last-modified date, although these times differ by up to 20 minutes across devices. In one instance, there was an incomplete entry in the [c]results.txt[/c] file: [QUOTE]UID: ixfd64/xxxxx, no factor for Mxxxxxxxx from 2^73 to 2^74 [[/QUOTE] (note the trailing bracket) This suggests mfaktc suffered some sort of I/O error. The terminal history doesn't show anything usual. It's probably also not a disk space issue because [C]/dev/sda1[/C] is less than half full. I'm at a loss to explain what might have happened. Anyone got ideas? |
| All times are UTC. The time now is 15:24. |
Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2023, Jelsoft Enterprises Ltd.