climateprediction.net home page
Posts by old_user18828

Posts by old_user18828

1) Message boards : climateprediction.net Science : Any way to resume a project after crash? (Message 25293)
Posted 25 Nov 2006 by old_user18828
Post:
Andrew

When a model crashes, the server flags that dataset for possible re-issue. (Up to a total of 4 re-issues.)
Whether or not it DOES get re-issued is another matter, as there are millions of combinations of values for the parameters.
You can see if a model is new or a re-issue by looking at the last digit in the name. If it\'s zero, it\'s a previously used set.
You can also see who has previously tried to crunch that dataset by looking at the Work unit ID, as displayed on the page listing all of your models.

Also, when the program gets written and tested, it will be possible to start part way through a failed model, from restart dumps which are uploaded every 40 years.
These, no doudt, will also be issued at random, although at this point, only the most basic of information is known about how it will work.




I wish I\'d seen this before agonising about a nearly finished model that got lost in mysterious circumstances. Have now restarted an earlier one that had been in suspended animation for a oouple of months... going fine so far....

There is so much material on the user forums (BOINC and CPDN) that it\'s easy to miss something quite fundamental like this. Maybe it\'s in one of the \"sticky\'s\"? Must have a look.
2) Questions and Answers : Windows : How to stop getting new Work Units? (Message 25276)
Posted 24 Nov 2006 by old_user18828
Post:
The BOINC client needs to know where the file is. There are several edits that need to be made to the client_state.xml file so that the client will know where the files are. I am sorry I don\'t have all the details. I have always just abandoned any models in similar circumstances, it hasn\'t seemed worth the risk to the other projects to try to recover to me.


Thanks for that advice, that\'s what I\'ll do too. Editing the files isn\'t something I feel it would be wise for me to attempt!

It would have been useful to know why the project file disappeared from the BOINC folder on this occasion, to avoid it happening again, but I\'ll just have to be more careful.
3) Questions and Answers : Windows : How to stop getting new Work Units? (Message 25234)
Posted 22 Nov 2006 by old_user18828
Post:
This is a further query about something I\'ve asked about in the message boards, but it looks as if this is a better place to discuss problems, so apologies for repetition.

I have BOINC set not to get new work just now, because I want it to pick up again on a WU (ResultId 5719246) that finished prematurely when 84% done. I have copied this unit back into the directory from back-up (thanks Mike Mars for advice), but when I did so found that there was no existing project file, although there were several others that have failed to complete previously. I copied in the back-up just after upgrading to version 5.4.11 (using the Windows \"Install/Uninstall\" procedure in Control Panel)and wonder if I somehow managed to get rid of this project file at the same time?

At any rate, although the project file IS now in the BOINC directory, it doesn\'t appear in the BOINC manager screen. I tried unchecking \"Read Only\" in the folder file Properties in Windows Explorer, as recommended in another post, but it has made no difference. Somewhere a while ago I came across a discussion of when a project is truly defunct and incapable of being resurrected, but I can\'t find it now. In \"My account\" the server state for this WU is \"over\" - does this mean it can\'t be got going again?

I have another 34% complete model (type hadcm3lbm),currently suspended. Should I give up on ResultId 5719246, the 84% one, and revert to the 34% one, or allow the project to give me an entirely new WU? I\'m not sure which is the best model to be running at the moment but gather from some of the discussion that some are more useful than others. Sorry to keep going on about this - I\'d really like to start getting some successful models again! Thanks!




©2024 climateprediction.net