I had a view unexpected errors with no clear reason. I suppose it has to do with memory access violation.

The only error on my results-list still there is: https://www.worldcommunitygrid.org/ms/device/...og.do?resultId=1855097844

If I remember correctly at that time BOINC just started a 4-core ATLAS from LHC@home . . . coïncidence?

SSD has enough space.
RAM: 8 GB, DDR3, non-ECC, has undergone Memtest recently for a different reason. Test was okay.
The mainboard is an old Fujitsu D3041, Chipset Intel G41
Processor is an old Intel Core2 Quad Q9550 @ 2.83GHz, no overclocking.

So maybe this old system is not the optimal one for ARP ?

oh, sorry, I was not aware that the page I set the link to cannot be seen by others.
Anyway, the Error links shows the following:

Result Name: ARP1_ 0027300_ 086_ 0--
<core_client_version>7.14.3</core_client_version>
<![CDATA[
<message>
couldn't start app: Can't get shared memory segment name: shmget() failed</message>
]]>

what makes we wonder though is that is says "couldn't start app ..."
and still the task ran for 35 hours and the other one for 28 hours.
For me, this does not fit together, does it?

erich56

If those tasks really used that much CPU time doing something constructive, that's a very strange "log file." It looks like the sort of report one gets when the BOINC run-time needs to report an issue during set-up (no startup information, et cetera)...

I presume you're running some flavour of Windows, as that seems to be the place that throws up errors of that type (and not just for ARP - my Linux boxes keep getting retries for all projects where the single original task was on Windows and failed!

There have been various discussions about that issue. I haven't followed them closely (no Windows systems!) but I seem to recall that it was a mixture of client version and the number of shared memory segments already allocated to processes - users with plenty of memory spare were getting the error (which implies a table size limit somewhere was being hit.)

Given that, Mike's "what else was it doing at the time" query is likely to be pertinent.

But I'll return to my original remark - if it had managed to start the APR1 application properly I'd've expected to see at least the initial INFO lines and the "Starting WRFMain" line. Does the BOINC log on your system indicate that the workunit(s) in question ever checkpointed - in fact, checking said log for all lines containing the relevant work unit names might be revealing!

Good luck troubleshooting - hopefully all will become clear at some point.

Cheers - Al.

Unfortunately WCG's branded client offered on this site (at least for Windows) is still on that version and they can configure the URL to check against for updates to have it not report the existence of a newer version until they give the green light themselves on their servers. Fortunately, erich56, you can download a newer version here and install right over the old version and the work units in progress won't be lost, though they will revert back to the last point where progress was saved (checkpoint).