Please do not rely on time measurements to detect faked returns or invalid
results. Please create two tasks and run them through a validator.
jm7
Travis Desell
<[email protected]
u> To
Sent by: Richard Haselgrove
boinc_dev-bounces <[email protected]>
@ssl.berkeley.edu cc
[email protected]
Subject
09/03/2009 10:26 Re: [boinc_dev] result.cpu_time for
AM GPU applications
It would be really helpful to be able to know the elapsed_time of a
result, mainly to see exactly how fast these workunits are crunching
(and if anyone is trying to fake results). Maybe add something like
an elapsed_time or gpu_time field to the RESULT struct. Or maybe
just change cpu_time to compute_time and that way it would be valid
for whatever kind of workunit.
--Travis
On Sep 3, 2009, at 10:22 AM, Richard Haselgrove wrote:
> Travis Desell wrote:
>
>
>> Is there another variable to determine the time a workunit has spend
>> on the GPU? For whatever reason our CUDA applications are
>> reporting a
>> very small cpu_time value, I'm wondering if this is some issue we
>> have
>> in our application, or if the BOINC client is reporting it wrong...
>
> Recent versions of the client report both cpu_time and elapsed_time
> back to the server: this example demonstrates the multi-threaded CPU
> case, where CPU is greater than elapsed. You are interested in the
> related but opposite case, where for GPUs 'elapsed' is more
> interesting than CPU.
>
> <result>
> <name>06jul09-160-6M-64-a_17_18_0</name>
> <final_cpu_time>205491.000000</final_cpu_time>
> <final_elapsed_time>61600.312500</final_elapsed_time>
> ....
>
> Unfortunately, the default result table schema in
http://boinc.berkeley.edu/trac/browser/trunk/boinc/db/schema.sql
> still only contains a single field to receive cpu_time, so I assume
> the final_elapsed_time data is being discarded on receipt. Wouldn't
> it be better for projects, now that we are leaving the single-
> threaded CPU era behind us, if both time reports were stored by
> default?
>
----------------------------------------------------------------------------------------------------------
Travis Desell
<deselt @ cs.rpi.edu>
1-518-867-1054
Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ )
milky...@home ( http://milkyway.cs.rpi.edu/ )
Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA
----------------------------------------------------------------------------------------------------------
_______________________________________________
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.
_______________________________________________
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.
