Interested to know if anyone else is getting this with the short models. I noticed disk usage seemed to be getting a bit high for BOINC and on checking, the last 4 short models hadn't cleaned up after themselves though they had sent all zips and cleared from Tasks In Progress view. If others have had this is it only on nix boxen or a global issue?

Edit: Just to be completely clear, this is not crashed tasks leaving their detritus on my disk which I know is a problem but models that have finished without a hitch other than having to wait to report/upload zips on some occasions.

Just to confirm Eirik Redd's earlier reply to your post...

I run CPDN on two Ubuntu machines, and I don't think any of the HADCM3S tasks that downloaded successfully have ever cleared up after themselves - it's been a regular task to recover the disc space!

I'm seeing this with both BOINC 7.2.33 (on 12.04) and BOINC 7.4.8 (on 14.04) which I got from costamagnagianfranco's PPA. So if it is the client rather than the application causing the clean-up problem, it's in the latest release candidate as well as the current production version.

Hopefully someone will chip in with something definitive about why it might be happening (or perhaps we need a separate thread to motivate that?) The current behaviour is certainly a nuisance!

Al.

---

Thanks, that's what I needed to know. i.e. it is nothing to do with my machine.

Curiosity makes me ask whether it happens on windows & Mac machines as well?

Hi Dave

I'm afraid that "leave stuff behind" thing is a problem with the model type.
I'm hopeful that some more testing will get done on these with a new compile to clear up some of the things that got rushed through testing.

I am getting different errors for my short runs.
Stderr: http://climateapps2.oerc.ox.ac.uk/cpdnboinc/result.php?resultid=17125779

I am also getting a windows error dialog (Visual Fortran run-time error):

forrtl: severe (17): syntax error in NAMELIST input, unit 5, file
E:\...\projects\climateprediction.net\hadcm3s_2
nvt_1981_2_009047449\jobs\climate.cpdc, line 393, position 19

followed by a stack trace.

I am getting different errors for my short runs.
Stderr: http://climateapps2.oerc.ox.ac.uk/cpdnboinc/result.php?resultid=17125779

I am also getting a windows error dialog (Visual Fortran run-time error):

forrtl: severe (17): syntax error in NAMELIST input, unit 5, file
E:\...\projects\climateprediction.net\hadcm3s_2
nvt_1981_2_009047449\jobs\climate.cpdc, line 393, position 19

followed by a stack trace.

... as I understand it, the Visual Fortran message appears only if the BOINC installation is not a service. For service installs the model still fails but the message box is suppressed. There is a small anomaly close to the reported location in the NAMELIST file (tab instead of space) but a compliant FORTRAN compiler would not, as far as I know, treat that as an error. So the error message appears to be a dead end.

Just got 4 more errors on the 'short' application.

Just got 4 more errors on the 'short' application.

From the new batch or not?

There was some talk of more testing which hoped to stop this for the next release. Also waiting to see if new batch clean up after themselves.

Short models created today still show download errors:

<core_client_version>6.10.43</core_client_version>
<![CDATA[
<message>
app_version download error: couldn't get input files:
<file_xfer_error>
<file_name>hadcm3s_7.24_windows_intelx86.exe</file_name>
<error_code>-120</error_code>
<error_message>signature verification error</error_message>
</file_xfer_error>
</message>
]]>

No, that's not a problem with models created today - that's the application file, unchanged since 23 July 2014.

Signature errors like that are usually the result of an anti-virus program or some other malware blocker interfering with the download. Some people reported success with that same file, using the technique I described in message 50291.

Yep just got 4 errors myself, and downloaded another 4 'short' and they are doing the same thing.

---

It's not a problem on Linux.

The new short tasks are also giving the INVALID THETA error on my Win XP64 machine. It has never finished any short tasks successfully. I have now disabled short tasks for that machine. Below is stderr from one of the latest tasks.

<core_client_version>6.12.34</core_client_version>
<![CDATA[
<message>
The device does not recognize the command. (0x16) - exit code 22 (0x16)
</message>
<stderr_txt>

Model crashed: ATM_DYN : INVALID THETA DETECTED.                                                                                                                                                                                                                               tmp/pipe_dummy                                                                  2048    

Model crashed: ATM_DYN : INVALID THETA DETECTED.                                                                                                                                                                                                                               tmp/pipe_dummy                                                                  2048    

Model crashed: ATM_DYN : INVALID THETA DETECTED.                                                                                                                                                                                                                               tmp/pipe_dummy                                                                  2048    

Model crashed: ATM_DYN : INVALID THETA DETECTED.                                                                                                                                                                                                                               tmp/pipe_dummy                                                                  2048    

Model crashed: ATM_DYN : INVALID THETA DETECTED.                                                                                                                                                                                                                               tmp/pipe_dummy                                                                  2048    

Model crashed: ATM_DYN : INVALID THETA DETECTED.                                                                                                                                                                                                                               tmp/pipe_dummy                                                                  2048    
Sorry, too many model crashes! :-(
07:22:24 (2292): called boinc_finish

</stderr_txt>
]]>

---

Harri, I see that all you wingmen also seemed to fail with their bits of those work units. They were also all using windows. I still have a couple of the old batch to complete before knowing whether the new ones complete on Linux or not.

INVALID THETA means that the model's physics has become unstable.
Which is what the researchers are looking for. All perfectly normal.

If there's a lot of them, then the values for the forcing parameters used for each of the models must be pushing the model's physics close to the limits of stability. This is called research.

And only the researcher knows what he's trying to find out at the moment.

I recognise that Les but all of the ones I looked at fell over at about 5 seconds which makes me wonder if maybe there is something else going on.

Sorry Dave. That was meant for Harri.

I didn't look too far into failures.
My first time models start with "30", so the new series may be 30, 31, 32, etc.
They're also for 2003. And should finish soon.

My other machine picked up 4 rejects, 5 if you count a "permanent download failure".
Not sure how far these will run.

So, a mix of old/bad and new/good are floating around.

I have had 3 of these short models fail with

Model crashed: ATM_DYN : INVALID THETA DETECTED

I presume we will get some credit(s) for the work done. Also I hope there is some scientific value gained.

---

<img border="0" src="http://boinc.mundayweb.com/one/stats.php?userID=343" />

None of the UK Met Office HadCM3 short v7.24 ever terminated successfully on my iMac : http://climateapps2.oerc.ox.ac.uk/cpdnboinc/results.php?hostid=1314566

On the PC the situation is not brilliant but still better : http://climateapps2.oerc.ox.ac.uk/cpdnboinc/results.php?hostid=1304501

On the PC the situation is not brilliant but still better :

And as those who have followed all of this and other threads on the subject, many pc's have not been able to complete a single task of this type. whereas most linux boxes seem to finish them all, certainly they seem pretty bomb proof on my machine so far, even surviving power failures without problems.

Credits are awarded per Trickle received at the server. If your tasks sent Trickles, you will get credit eventually (not instantaneously). Of course, there is typically a period between last Trickle and crash point that won't be compensated.

---

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.

I'm not worried about credits and I know about the trickle mechanisms that's used here (and it's great !), it's more that I wonder "what the heck with this application" :D