My model fails at 35.92% (at the end of 24/5/1978). This is happening consistently: 3 times I’ve restored from different backups, rerun, and it’s failed at the same model time – so I deduce it’s something to do with the model, not clashes with other processes. I’m running BOINC Manager 5.8.16. The error messages are sparse:

2007-08-10 16:36:27 [climateprediction.net] Restarting task hadcm3inct_cn23_1920_160_05890640_1 using hadcm3i version 540
2007-08-10 17:10:29 [climateprediction.net] Deferring communication for 1 min 0 sec
2007-08-10 17:10:29 [climateprediction.net] Reason: Unrecoverable error for result hadcm3inct_cn23_1920_160_05890640_1 (The device does not recognize the command. (0x16) - exit code 22 (0x16))
2007-08-10 17:10:29 [climateprediction.net] Computation for task hadcm3inct_cn23_1920_160_05890640_1 finished

Searching the forum, there are some references to exit code 22, and a mention of a fix on its way. Or might it be worth restoring from an earlier backup? Or should I cut my losses, and pick up another model? (This is my second one. I’m familiar with the good practice guidelines on running a model, but no expert at technical error diagnosis. )

Incidentally, I’d noticed that the graphics showed a very cloud-free earth – don’t know whether that implies that this model is unsustainable anyway.

Your advice would be appreciated!

Charlotte

Charlotte,

The newer version of the science application does have a fix for some of the problems that cause models to end with an error code 22. You have maintained a nice clean error record on the results page (here), so I assume that network activity is turned off. I suspect that the only way forward is to turn the network activity on, at which point a more intelligible error might appear (\'negative pressure\', \'negative theta\' etc.). After three restores, I doubt whether the model can be saved, even by an earlier backup.

The models appearing now seem to be shorter (80 years rather than 160 years) or you can select a \'slab\' model in your preferences. A bit more variety than formerly ...

Iain

to turn the network activity on, at which point a more intelligible error might appear (\'negative pressure\', \'negative theta\' etc.)Iain

These errors are written to the stderr.txt file in the BOINC project folder anyway aren\'t they? In the slots/0 sub-folder (that\'s were it is in mine anyway). Might be worth Charlotte having a look, though no doubt you\'re right about it not being recoverable.

---

Model crashed: umshell1.f: TRANSO2A: Missing data in ocean UV fields

That\'s what was being masked by the \'22\', and isn\'t recoverable, so best to start again with an \'improved\' model.

---

Backups: Here

Thank you Iain, GavinK & Les for your responses.

I switched network activity back on, then saw the output on the results page. Just one question: did the \"Missing data in ocean UV fields\" arise from something I (or my computer) did, or was is a natural consequence of the model parameters? To put it another way, could I have avoided it?

My computer is slow at running these things (4.19 s/ts average) and the full 160 model years lasts 8 or 9 months real time, so I\'ll now set my Preferences for a Slab model (after updating to the latest BOINC version).

Many thanks for all the helpful information you (and all the others) provide.

Charlotte

It was a model problem.

---

Backups: Here

My model fails..

I’m running BOINC Manager 5.8.16.
2007-08-10 17:10:29 [climateprediction.net] Reason: Unrecoverable error for result hadcm3inct_cn23_1920_160_05890640_1 (The device does not recognize the command. (0x16) - exit code 22 (0x16))
2007-08-10 17:10:29 [climateprediction.net] Computation for task hadcm3inct_cn23_1920_160_05890640_1 finished

(I’m familiar with the good practice guidelines on running a model, but no expert at technical error diagnosis. )

Charlotte

Very similar situation here.

Here is my relevant CPDN error log:

31/08/2007 10:35:07 AM|climateprediction.net|Deferring communication for 1 min 0 sec
31/08/2007 10:35:07 AM|climateprediction.net|Reason: Unrecoverable error for result hadcm3inct_ckvb_1920_160_05877532_3 (The device does not recognize the command. (0x16) - exit code 22 (0x16))
31/08/2007 10:35:07 AM|climateprediction.net|Computation for task hadcm3inct_ckvb_1920_160_05877532_3 finished
31/08/2007 10:35:07 AM|climateprediction.net|Output file hadcm3inct_ckvb_1920_160_05877532_3_12.zip for task hadcm3inct_ckvb_1920_160_05877532_3 absent
31/08/2007 10:35:07 AM|climateprediction.net|Output file hadcm3inct_ckvb_1920_160_05877532_3_13.zip for task hadcm3inct_ckvb_1920_160_05877532_3 absent
31/08/2007 10:35:07 AM|climateprediction.net|Output file hadcm3inct_ckvb_1920_160_05877532_3_14.zip for task hadcm3inct_ckvb_1920_160_05877532_3 absent
31/08/2007 10:35:07 AM|climateprediction.net|Output file hadcm3inct_ckvb_1920_160_05877532_3_15.zip for task hadcm3inct_ckvb_1920_160_05877532_3 absent
31/08/2007 10:35:07 AM|climateprediction.net|Output file hadcm3inct_ckvb_1920_160_05877532_3_16.zip for task hadcm3inct_ckvb_1920_160_05877532_3 absent


A while back, I was running 2 different CPDN tasks, 1 on each core of my dual core (Intel C2D E6400) CPU.

Then the 1st task crashed somehow, so I gave up on it after browsing these forums a little and being unable to work out if in fact the processing I had done to date would actually be any use to the project.

Now with this recent crash of my 2nd CPDN task (per logs above), I am similarly unsure weather all of the (significant) CPU time I have applied to this has at all been any real use.

Can someone please at least clrify weather or not any of my work done on these incomplete models will be of any use?

---

ssar,

Some models are doomed to fail by the set of parameters they start with, but most models can be made to finish, though in practice that may mean taking a backup to get around a PC-induced crash. The first model got to 2006 and looks as if it was a PC-induced failure; the second model got to 2030 and looks to be of the doomed variety (it ends with repeated \'Missing data in ocean UV fields\').

Minor uploads are made every model year, larger ones every decade and a full restart dump every forty years - so your results will have got back to the project. A lot of the scientific work has been directed towards the period up to 2000, which both your models cover. So, yes, they are useful.

To increase the probability of finishing, it\'s well worth taking a backup. Or you can change your account preferences to download a shorter \'slab\' model (45 years). The standard coupled models are also shorter at the moment (80 years, rather than 160 years).

If you look at Les\'s post, two before this one, there are some useful links there on avoiding crashes and making backups.

Iain