See the ACTIVE_TASK::handle_premature_exit() function in app_control.cpp for
the actual implementation. It's 100 rather than the 10 you propose, so
theoretically about 53 minutes (100*~32 seconds). I have seen a few Task detail
pages showing the "too many exit(0)s" error message.
If that were to be reduced, I think the core client should have some kind of
check whether it has actually sent a heartbeat to the application recently.
Also it might make sense for the temporary exit capability to be given a larger
count or maybe even exempted.
--
Joe
On Fri, 05 Aug 2011 08:33:27 -0400, <[email protected]> wrote:
> I had a fix for #3 once upon a time. If a task resets to the very same
> check point ten times in a row, it should be declared dead. Note that if
> it moves on to the next checkpoint, the count should be restarted.
>
> Not having this check is a way for wasting days of CPU time.
>
> jm7
>
>
> robert miles
> <robertmiles@bell
> south.net> To
> Sent by: <[email protected]>
> <boinc_dev-bounce cc
> [email protected]
> u> Subject
> Re: [boinc_dev] 0% progress for
> Pentium(R) Dual-Core CPU and
> 08/04/2011 07:57 Windows 7 combination
> PM
>
>
>
>
> I've seen a few fairly similar problems you and the users may want to
> check for:
>
> 1. The workunit crashes in such a way that it stops using any CPU
> time at all, and does NOT tell the BOINC client this has happened.
> It may or may not tell Windows something has gone wrong. This
> often freezes the heartbeat tests so it may even go well past the time
> when it should have halted due to running too long.
>
> 2. A few programs give different results depending on whether
> they run on Intel CPUs or AMD CPUs.
>
> 3. Some workunits repeatedly end without an error code, but
> also without producing the output file to tell the BOINC client
> that it finished properly. Should be visible in at least some of
> the log files.
>
> 4. Are the checkpoints closely spaced enough that those users
> should be able to reach them even with the following combination
> of settings?
>
> a. The usual 60 minutes before possibly going on to some
> other BOINC workunit from a different project.
>
> b. Only 60% of the CPU time available due to running on
> a laptop.
>
> c. Enough other BOINC projects enabled with enough
> resource share that if the 60 minutes is reached, the
> workunit will usually be suspended.
>
> d. Either the option to keep suspended workunits in
> memory is not enabled, or BOINC is not allowed to use
> enough memory to do it.
>
> 5. A few workunits do not have the progress reporting
> working properly, but go on to produce correct results
> otherwise.
>
> 6. Is there any difference between running with graphics
> enabled and running without graphics enabled (if you even
> have a screensaver section yet)?
>
> 7. Does recompiling the application specifically for that
> CPU and OS combination offer any improvement on the
> machines with the problem?
>
> Robert Miles
>
>
>
>> Date: Thu, 4 Aug 2011 10:07:47 -0400
>> From: Boyu Zhang <[email protected]>
>> Subject: [boinc_dev] 0% progress for Pentium(R) Dual-Core CPU and
> > Windows 7 combination
>
>> Hi All,
>
>> We have a boinc project distributing from our lab, and I observed from
> the
>> database that there is a high percentage of hosts with CPU Pentium(R)
>> Dual-Core CPU and operating system Windows 7 has the same problem. The
>> problem is that the boinc application running on the hosts constantly has
> 0%
>> progress no matter how long it runs. But on Linux system we don't see the
>> same problem.
>
>> Moreover, some hosts with exact the same cpu and os version, but some of
>> them are successful while others are having the problem, we tried to
>> reproduce the situation on the machines in the lab but with no luck so
> far.
>> Did anyone have the same problem? Any help and suggestions is
> appreciated,
>> thank you!
>
>> Boyu
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
