climateprediction.net (CPDN) home page
Thread 'Comment re: % complete reset @ phase changes'

Thread 'Comment re: % complete reset @ phase changes'

Message boards : Number crunching : Comment re: % complete reset @ phase changes
Message board moderation

To post messages, you must log in.

AuthorMessage
Profileold_user11965

Send message
Joined: 4 Sep 04
Posts: 61
Credit: 80,585
RAC: 0
Message 13109 - Posted: 4 Jun 2005, 3:54:03 UTC

I\'ve noticed that recently downloaded models reset the % completed field at each phase change. Not that it matters to the science, but I rather dislike this. I prefer 100% to be 100%, not 33.333% or 66.666% or 100% depending on the phase.

Not that it\'s the end of the world or anything. Just my opinion.
ID: 13109 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 13112 - Posted: 4 Jun 2005, 6:45:56 UTC

There was a re-build of the server software a few days ago. Perhaps that.
What version of BOINC are you using, and what version of hadsm?

I'm due for a new model in 6 days, and I'd like to be prepared for odd behaviour.

Les

ID: 13112 · Report as offensive     Reply Quote
Profileold_user11965

Send message
Joined: 4 Sep 04
Posts: 61
Credit: 80,585
RAC: 0
Message 13157 - Posted: 6 Jun 2005, 9:32:34 UTC - in response to Message 13112.  
Last modified: 6 Jun 2005, 9:32:48 UTC

> There was a re-build of the server software a few days ago. Perhaps that.
> What version of BOINC are you using, and what version of hadsm?
>
> I'm due for a new model in 6 days, and I'd like to be prepared for odd
> behaviour.

Hi, Les. I'm seeing the behaviour on hadsm3 4.13. It seems to have an interesting side effect that as the phase nears 100%, the core client downloads another WU, thinking that it'll soon be out of work. This happened under CC 4.43 Windows and CC 4.19 Linux. On the Linux box, I've upgraded to 4.43 so that the improved scheduler will have a chance to 'panic' and ensure that both WUs are completed in time. On the Windows box, I'm doubtful that any amount of panic will get the second WU done because the machine is a fairly old 800 MHz P-III.

I originally thought that it was a core client problem that caused the Windows system to download the new CPDN unit immediately after the upgrade to 4.43. That may indeed be the case, but it doesn't explain how my 4.19 core client downloaded a new unit as it neared the end of phase 1 on this model run. As such, I now consider this % resetting between phases to be a bonafied bug.

Cheers.
ID: 13157 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 13171 - Posted: 6 Jun 2005, 12:44:38 UTC

I think that the early download of an extra model is a bug in BOINC 4.43, and apparently fixed in 4.44, but it shouldn't happen in 4.19.
However, I've seen something similar happen to one person downgrading from 4.43 (I think), to 4.19. The problem from 4.43 followed him with the downgrade. It may depend on how it is done. Perhaps if the higer version isn't removed, just overwritten.

Strange things happening lately.

Les



ID: 13171 · Report as offensive     Reply Quote
Profileold_user11965

Send message
Joined: 4 Sep 04
Posts: 61
Credit: 80,585
RAC: 0
Message 13187 - Posted: 7 Jun 2005, 5:45:06 UTC

I should probably also point out that my Windows box is running hadsm3 4.12. I think the percentage done reset is new to v4.13.
ID: 13187 · Report as offensive     Reply Quote

Message boards : Number crunching : Comment re: % complete reset @ phase changes

©2024 cpdn.org