These are Jeremy\'s tasks.

I think the significant thing is that Jeremy has upgraded today from BOINC 5.10.45 to BOINC v.6 and may not have let BOINC itself migrate the model data correctly. So BOINC has lost sight of two models and is calling them \'detached\'.

Jeremy, can you tell us whether the following sequence of events is correct?

On Monday 12 Jan you had model 7674417 (half completed) and 7679952 (1/3 completed) running with BOINC 5.10.45. 7679952 crashed and the server immediately sent you 7725474 to replace it.

This morning Tuesday just before midday you upgraded to BOINC 6. BOINC 6 didn\'t find the two previous running models 7674417 and 7725474, so the CPDN server sent you two new models (the two at the top of your task list).

Sorry, I got one of those model numbers wrong earlier so I\'ve edited my post.

Is that time sequence correct?

We need to know

* Have you still got a backup of the complete BOINC 5.10.45 folder made after exiting from BOINC and before 7679952 crashed, or after it crashed but before you upgraded?

* Later edit: Could you please also tell us what backup method you used?

(I think it may be possible to rescue one or both of those part-completed models, but only if you made a backup before you upgraded BOINC.)

Very useful info, thanks. This means the model\'s progressed 2575 timesteps in more than 7 days which means about 360 timesteps per calendar day. You\'re going to have to put this defective model out of its misery by aborting it. Better luck with your next model and thanks for reporting the problem. I wish more members would post to report anomalies and abnormalities.

Before aborting it you can if you want in the CPDN preferences section of your account select the type of model you download next.

I\'ll send cbamber a private message.

Reminder to everyone. In your CPDN accounts please enable email notification of private messages so if we have to contact you about this sort of problem you\'re more likely to notice the message.

Hi again Jeremy

If the graphics globe looks uniformly blue for temps and other uniform colours when you press R for precipitation or P for pressure, the prognosis for this model is dire.

One of the other crunchers you mentioned aborted it. Cbamber is battling on but his/her model hasn\'t trickled for two weeks.

In my experience these slow-processing models can\'t be rescued and have to be aborted. One member, John Hunt, had this happen to a model near the end of a phase and he very kindly let it crunch on for ages so it would upload its end-of-phase zip file and we\'d see what sort of results it produced. From the moment of the slowdown the results suddenly became rubbish, unusable by the researchers. Look at John\'s phase 3 precipitation graph.

These model worlds don\'t really turn to ice, but the graphics look as if they do. It\'s really the processing that freezes up.

Just a couple of things for you to check please before you abort this model and one of us sends a private message to cbamber.

* Is your computer overclocked?

* What sec/timestep does your graphics window now say? (This figure won\'t have slowed to its true current snail\'s pace because it\'s a cumulative average, but if this really is an iceworld the sec/ts should already indicate this.)

Two possibilities:
1) It\'s turned into an \"ice world\" Info
2) You interrupted the model at the phase change and before it reached the 1st savepoint in the next phase. (No real help thread, just replies in various places, and warnings in News and Announcements.

My guess is 2) (No more trickles will be registered until the model catches up to where it was.)

Jeremy, you were surprised that your completed slab model disappeared from the Tasks window of your BOINC manager. This disappearance is normal when a task is reported. But it\'s still on your CPDN web pages. Here it is, with lovely graphs:

http://climateapps2.oucs.ox.ac.uk/cpdnboinc/result.php?resultid=6977062

If you select the type of model you want in future in your CPDN project preferences in your account, you won\'t be offered any more unsuitable models.

Let us know what you eventually decide and how you get on.

Let us know what you eventually decide and how you get on.

Yes I did create a new folder when I tried to restore.

Normally you wouldn\'t need to create a new BOINC folder when you restore a backup. Before restoring the backup you\'d double-click on the BOINC folder to display its contents, then select all the contents, then delete everything. Going back a page, the BOINC folder would still be there, but empty and ready to receive the contents of the backup folder.

But if you do for any reason delete the BOINC folder itself, making a new one shouldn\'t be a problem as long as you make it in the same location as before.

But if you created a new BOINC folder without deleting the contents of the old one, I don\'t think the restore would work. A computer should never have more than one instance of BOINC on it.

It was probably the \"save while BOINC was running\" that caused the problem.
Xcopy takes a long time to copy everthing, by which time some more files would have been written, making the \"set\" incompatable, and useless.

I expect you know that when you restore a BOINC folder backup, you have to paste it into an empty BOINC folder? You can\'t add the backup to stuff already in the BOINC folder.

I think that if I had that backup of a 160-year model which had already completed more than 20 model years, I\'d save the backup and at some point I\'d try to restore it again. Maybe even in 2009 after the current tasks have completed. But I\'d first check the workunit that the backed-up task belongs to to see whether anybody else\'s computer had run it to completion. If yes, I wouldn\'t spend time replicating another computer\'s efforts. If no, I\'d have a final shot at restoring it.

When I use Les\'s manual backup method, I find that if I\'ve ever forgotten to exit from BOINC first, some of the files in the BOINC folder won\'t copy to the backup folder and I have to start again.

Basically though, as I don\'t think you\'re letting the computer process models 24/7, you\'d be better off only running shorter HADSM slab models when you need new work. You can select what models you get in future in the CPDN preferences of your account.

Oops, my mistake there - I\'d forgotten what it was called. I agree that the simple manual backup method is very useful as one sees exactly what one is doing.

...Is there any way I canretrieve the lost data?

You appear to have two models running: 7340333 (HADCM3-coupled) and 6977062 (HADSM3-slab). The slab model is almost finished, so it may be a good idea to let that run to completion. Make sure the \'no new tasks\' button is pressed in BOINC Manager, so that you don\'t get any new models. The coupled model has only just started - you\'re in for a long haul with that one, since it\'s a 160-year model running (hyperthreaded) with the slab.

The other model you mention, 6685813, is also a coupled model but has made a bit more progress. It\'s reported as crashed, but there is a temporary problem on the Web site that is hiding the error messages - so we can\'t tell what caused the crash. It should be possible to restore it from a backup and continue.

If I were you, I would:

1) Press \'no new tasks\'.

2) Suspend the coupled model.

3) Let the slab model finish.

4) Resume the coupled model.

5) Forget about the crashed model.

However, that\'s just my opinion!

PS If the two computer records on your computer page (here) are actually one physical computer, then it\'s a good idea to \'merge\' them - see the \'merge this computer\' link at the bottom of the computer summary page for each computer.