Message boards : Projects : I'm looking for a project that respects checkpoints.
Message board moderation
Author | Message |
---|---|
Send message Joined: 19 Jun 23 Posts: 7 |
HI Added the following to the global_prefs.xml settings: <disk_interval>60</disk_interval> However, many projects seem to ignore checkpoints. I'm looking for a project that respects checkpoints. I'm using linux. best regards. |
Send message Joined: 17 Nov 16 Posts: 891 |
HI Depends on the science application mostly. Not the project per se. What kind of tasks are you interested in? CPU or GPU? Or both? |
Send message Joined: 19 Jun 23 Posts: 7 |
I am cpu only. |
Send message Joined: 19 Jun 23 Posts: 7 |
I found. NumberFields@home works as configured of checkout. Except for the NumberFields@home? |
Send message Joined: 25 May 09 Posts: 1302 |
You might find setting your checkpoint interval in global_prefs_override.xml will work better that using that set in global_prefs.xml , the latter can be written/over written by projects, while the former is under your control. |
Send message Joined: 17 Nov 16 Posts: 891 |
You might find setting your checkpoint interval in global_prefs_override.xml will work better that using that set in global_prefs.xml , the latter can be written/over written by projects, while the former is under your control. If you read his OP, you can see that he has already set his checkpoint interval to 60 seconds. Unfortunately, a majority of projects ignore this value and let their science applications determine when to checkpoint. Many cpu apps don't checkpoint at all. Even a lot of gpu apps don't checkpoint. With gpu tasks though, the penalty is less because of their short runtimes. Starting from scratch isn't as hurtful as losing all progress on a multi-hour or multi-day cpu task. |
Send message Joined: 25 May 09 Posts: 1302 |
Yes, but he's he's doing it in the "main" global_prefs file, which can be over-written by applications unlike the over-ride file which takes precedence. Bit of belt and braces. We are all forgetting that the value set by global_prefs is a minimum time period between an application doing a checkpoint, not an absolute value. If the application is in a critical point in its calculation a checkpoint may be skipped or extended. |
Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software Foundation.