Forum > Windows

Exceeded Elapsed Time Limit - Error??

(1/3) > >>

Ghost0210:
Hi,
As both the Seti boards are down at the moment, I'm hoping someone here may be able to help.  :D
I was halfway through processing a couple of Beta wu's, got to about 55% through the tasks andd they then both aborted with the following messages:
30/05/2010 19:52:26   SETI@home Beta Test   Aborting task 18dc09aa.22310.7429.5.13.83_0: exceeded elapsed time limit 272.626338
30/05/2010 19:52:26   SETI@home Beta Test   Aborting task 18dc09aa.22310.7429.5.13.82_0: exceeded elapsed time limit 272.626338
30/05/2010 19:52:28      [wfd] Request work fetch: application exited
30/05/2010 19:52:28      [wfd] Request work fetch: application exited
I've never seen a wu error out with this message before and was hoping that someone might either have seen it or be able to explain why the tasks self aborted
Thanks
 
[edit]Should have said hese tasks were cuda wu's - looks like I've had quite a few cuda tasks that have all aborted after about 4:30 run time

Claggy:
I've only seen Maximum Time Exceeded errors with Astropulse tasks,
eithier when the flops value in the app_info was set far too high,
or with Seti Beta's Stock Hybrid Astropulse app,
when a slow (AMD) CPU was paired with a fastish GPU,
since Hybrid Astropulse app is mostly a CPU app, with a bit of GPU computation thrown in,
slow AMD computers were having their tasks aborted,
because the project used the GPU flops value instead of CPU flops,

Was that 4mins 30secs, or 4hrs 30mins?, could be a VLAR task on a very slow GPU,
or the Cuda app might have dropped into CPU-fallback mode,

Claggy

Ghost0210:
Hi Claggy, thanks for the info think you may have sorted this for me :)
It was 4 minutes 30. All the tasks that seem to have had this error message have been running on one of my GT240's and only today have I noticed this happening.
All the cuda tasks that have run on my GTX260 have completed fine no matter what the run time, which is why I was confused
I do use a app_info on Beta though, and because there's such a difference in compute speed between the 240's and the 260 I have had to compromise on the <flops> value and try to go in the middle of the values
I guess that I may have been leaning to far with the flops towards the 260 and may then need to lower the value so the 240's get a more realistic est. runtime   ???

Ghost

Josef W. Segur:
The mb_splitter processes set the <rsc_fpops_bound> which controls the computation limit to ten times the <rsc_fpops_est> value. In the past that has been plenty, the host's Duration Correction Factor (DCF) is applied only to the estimated runtime and with most hosts having DCF of something like 0.2 that relative time ratio was more like 50 in practice.

The new credit system being tested at Seti Beta includes server-side code which scales the estimate and bound for work being sent to be done with each application on the host. The scaling is based on server statistics for the host's performance using that application. If it works as intended, the host DCF should tend to about 1.0 so that extra margin we had will be gone. And it's right that using the same application with two GPUs of widely differing capability is likely to cause problems. David Anderson may not have considered that since the BOINC default setting would only use the better GPU. The server of course doesn't have any way of knowing which GPU will actually be used, I'm not even sure the characteristics of anything other than the better GPU are sent to the servers.

The project might consider a larger margin in the bound set by the splitter. Meanwhile, reducing the <flops> setting should help at least for awhile though I suspect it may only cause the server-side adjustment to shift enough to have the problem come back after you've returned more results. It's good Beta testing of the new credit system in actual use...
                                                                                         Joe

Ghost0210:
Thanks Joe,
So it looks like this could be a mix of my fault :( for having a badly written app_info and the new credit system @ Beta
Anyway bit the bullet and got myself a new HD 5670 to replace one of the 240's, so removed the troublesome app_info.

Thanks for your answers, will have to remember next time I modify a app_info to write it for the slowest of my cards

Thanks
Ghost

Navigation

[0] Message Index

[#] Next page