Some notes on the changes.
- The switch to check_req true in the call to get_app_version() should indeed
resolve the first point, though the comment block for get_app_version() should
also be revised. The change will help somewhat with VLARs being expired, a host
with GPU may sometimes only ask for CPU work.
- Adding the explicit wu_is_infeasible_custom() call is redundant, that has
been called in wu_is_infeasible_fast() since late July 2009.
- I'm disappointed you didn't take the opportunity to make use of
result.app_version_id to try to resend for the same app_version as the original
assignment chose. That should be feasible in the vast majority of cases, and
users could easily understand the logic.
--
Joe
On Thu, 08 Nov 2012 02:44:06 -0500, David Anderson <[email protected]>
wrote:
> I checked in these changes.
> (Jeff, please deploy and test these on SETI@home beta at some point).
> -- David
>
> On 07-Nov-2012 7:09 PM, David Anderson wrote:
>> The current code resends lost jobs using the fastest app version,
>> whether or not work is being requested for the devices it uses.
>> There are 2 problems with this:
>>
>> - It can send work for devices that don't need it
>> - It can send VLAR jobs to GPUs
>>
>> I'll check in changes to prevent these behaviors.
>> May take me a day or two.
>>
>> -- David
>>
>> On 06-Nov-2012 4:37 PM, Stephen Maclagan wrote:
>>> I get the impression that the present scheduler is broken in some
>>> respect, on both my i7-2600K/GTX460/HD7770 and my E8500/GTX9800+ hosts
>>> they will request work for one device, then get sent work for another
>>> device that wasn't requesting work:
>>>
>>>
>>>
>>> 01/11/2012 19:02:57 SETI@home Beta Test Reporting 2 completed tasks,
>>> requesting new tasks for CPU and GPU
>>>
>>> 01/11/2012 19:02:57 SETI@home Beta Test [sched_op_debug] CPU work
>>> request: 64.03 seconds; 0.00 CPUs
>>>
>>> 01/11/2012 19:02:57 SETI@home Beta Test [sched_op_debug] NVIDIA GPU
>>> work request: 0.00 seconds; 0.00 GPUs
>>>
>>> 01/11/2012 19:02:57 SETI@home Beta Test [sched_op_debug] ATI GPU work
>>> request: 3768.39 seconds; 1.00 GPUs
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test Scheduler request completed:
>>> got 4 new tasks
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug] Server
>>> version 701
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test Message from server: Resent
>>> lost task 05ap10al.26099.13160.8.14.143_1
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test Message from server: Resent
>>> lost task 05ap10al.12468.72.9.14.67_0
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test Message from server: Resent
>>> lost task 05ap10al.12468.72.9.14.68_0
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test Message from server: Resent
>>> lost task 05ap10al.12468.72.9.14.69_0
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test Project requested delay of 7
>>> seconds
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug] estimated
>>> total CPU job duration: 0 seconds
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug] estimated
>>> total NVIDIA GPU job duration: 3306 seconds
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug] estimated
>>> total ATI GPU job duration: 0 seconds
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug]
>>> handle_scheduler_reply(): got ack for result 05ap10al.
>>> 26207.12342.140733193388040.14.220_1
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug]
>>> handle_scheduler_reply(): got ack for result 05ap10al.
>>> 26207.12751.140733193388040.14.183_1
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug] Deferring
>>> communication for 7 sec
>>>
>>> 01/11/2012 19:05:32 SETI@home Beta Test [sched_op_debug] Reason:
>>> requested by project
>>>
>>>
>>>
>>> I did initially report it on Seti Beta:
>>>
>>>
>>> http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=1950&postid=44291
>>>
>>>
>>>
>>> Turning off Use Nvidia GPU in the project preferences stops work being
>>> resent to the Nvidia GPU when it wasn't asking for it,
>>> i can then get CPU work resent to the ATI GPU when it wasn't
>>> asking for it instead,
>>>
>>>
>>>
>>> Claggy
_______________________________________________
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.
