What happened?

The work unit exited after 16/17 trickles and the client downloaded a new work unit.

[core_client_version]4.13
[message] - exit code -5 (0xfffffffb)
[active_task_state] 1
[signal] 0

---

Hi Steve

I recently upgraded to 4.13 and have had 2 models already crash out with this same -5 exit code. I believe that Carl said on another thread that this means that the model has produced negative pressure, which should not happen. It is a computation error. A window asked me whether I wanted to debug the programme and when I said Yes, up came a debugging window in C++, with a yellow arrow pointing to the line where the computation error presumably lay. As you can imagine, I was not going to start altering the lines of code! It doesn't seem able to debug itself.

As these models have gone wrong and I cannot put them right, I am not bothering to back up and attempt to restart them. When I had 4.04, I also had repeated problems with models stopping (for many different reasons)and new models downloading. On yet another thread, Carl told someone having similar problems that boinc cpdn does not get on very well with the user's particular type of Athlon processor. He advised the user to crunch the cpdn classic model instead.

I have a different Athlon processor, but cpdn boinc doesn't seem to like mine either. This is in spite of checking all my settings, and learning to start up and suspend/exit the models cleanly. I am fed up with crashed models on boinc, and am therefore moving back to cpdn classic, which worked beautifully on my machine for a whole year.

If you have repeated crashed models, you might consider doing the same.

If, however, this crash is just a one-off and your boinc models usually complete properly, you should stick with boinc, making sure you back the models up regularly so that you can try to get them going again. But in the case of a -5 exit code, I would just abandon the model because this code means it has gone wrong.

I would be most grateful if some more experienced and knowledgeable crunchers could add their comments because I know that I'm not the only cruncher in this position. Have I reached sensible conclusions?

---

__________________________________________________

The -5 error error is a model crash. If you are experiencing this problem with the AMD processors, then you can revert to the classic client.
this could be due to the different fortran compilers used for BOINC and classic i,e Intel and Compaq respectively.

---

WIn 98 not listed as being supported to run CPDN under BOINC.


http://climateapps2.oucs.ox.ac.uk/cpdnboinc/tech_faq_boinc.php

---

<a href=\"http://boinc-doc.net/boinc-wiki/index.php?title=Main_Page\">Link to Unofficial Wiki for BOINC, by Paul and Friends</a>

&gt; WIn 98 not listed as being supported to run CPDN under BOINC.

It seems that <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=948">some people</a> are able to run it happily after a few tweaks from the team, but AMD processors seem always to have given more problems running CPDN under Win9x, even in classic.

An Intel processor running HT under Linux still seems to be the platform of choice for this project!

---

&gt; If you have repeated crashed models, you might consider doing the same.

In this case, the PC has an Intel PIII-866MHz CPU....so the Athlon aspect won't apply.

&gt; If, however, this crash is just a one-off and your boinc models usually
&gt; complete properly, you should stick with boinc, making sure you back the
&gt; models up regularly so that you can try to get them going again. But in the
&gt; case of a -5 exit code, I would just abandon the model because this code means
&gt; it has gone wrong.

It was the first workunit for CPDN on that PC. Prior to that, it had been happily crunching SETI work units for 3-4 years, 24/7.....and still does 50% of the time.

Thanks for the response.

Steve

---

&gt; WIn 98 not listed as being supported to run CPDN under BOINC.

It was supposed to be OK as of v4.12 or 4.13.....

I received e-mail from CPDN about this.

---

&gt; &gt; WIn 98 not listed as being supported to run CPDN under BOINC.
&gt;
&gt; It seems that <a> href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=948"&gt;some
&gt; people</a> are able to run it happily after a few tweaks from the team, but
&gt; AMD processors seem always to have given more problems running CPDN under
&gt; Win9x, even in classic.

That may depend on whether you overclock the CPU or not. It may depend on the quality of your RAM, too.

Most of my systems (6 of 7) run AMD Athlon CPUs - not overclocked - and execute the models flawlessly. The system with the problem is Win98SE on an Intel PIII-866CPU.

&gt; An Intel processor running HT under Linux still seems to be the platform of
&gt; choice for this project!

Most of my systems (5 of 7) run Linux.....so I agree there. But my own experience says AMD (not overclocked) aren't a problem.

---

Hi all

In my case, I have Win2000 not 98, so it has to be the Athlon that's the problem.

Every model I tried with boinc 4.04 and 4.13 would produce (before crashing) temps from May to Sep 1811&amp;12 that showed white patches ie off the top of the temp scale, which is not what a model should produce for those years. Sometimes there was white from the W Sahara to Pakistan. All those models, and there were quite a few, couldn't have possibly have had the same dodgy parameters. So this had to be how my machine mis-processed the data. No wonder they all crashed.

Having returned to classic, the present model is showing the summer temps you would expect for phase 1. So as the machine is not o/c, is running cool and is working beautifully in every other respect, I will just stick with this machine and classic until it breaks down. At that point I'll make sure I get a boinc-compatible processor.

The only problem about being stuck on classic will be missing out on the new interesting experiments.

---

__________________________________________________

When I was puzzling over the debugging C++ page, there seemed to be lots of abbreviated verbs, like pus for push etc. Plus frequent instances of mov which for 9 seconds I thought was my name, just erroneously missing the full stop.

---

__________________________________________________

I'm quite sure its is a conzeptional bug in BOINC - it happens only with CPDN as CPDN consists of multiple programs.

- BOINC starts program A
- program A starts program B
- BOINC shuts down and kills program A
- program B survives and holds files open.

I can easily reproduce this -5 error by starting BOINC on a Dual CPU machine, then starting two Seti-Classic tasks or running a WinAmp video to create load on the CPUs. Exiting BOINC always leaves two "living" CPDN tasks behind then - if you kill those with a task manager immediately, CPDN will restart properly but if you do not care about them, both models will crash and the projects error output will contain messages about files that could not be opened or created.

It does not happen if the machines are only lightly loaded.

I had it on a Dual P3/1266 and on a Dual MP2600+ so far, both running Win2k.


The fix has to be done on BOINC side, one possible solution would be a list of secundary PIDs that a client has started so BOINC could shut them down in the proper order.

A cleaner way would be if BOINC would announce shutdown and then wait for a "ready for shutdown" signal from the project client before it kills it.