Mind you, the explanations offered by both Claggy and Holmis only apply to Gravity Wave (CasA) tasks - they're the only ones which use the locality scheduler.
The workunits reported were - properly for this thread - BRP jobs, which download fresh data with every task. I suspect the real explanation was more prosaic - replacement tasks are put at the back of the queue, and with things being quiet here until testing resumed this morning, there were probably very few active computers plodding their way through that queue.
Anyway, all WUs have been completed and validated now.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
Recently I got a bunch of BRP v1.33 tasks (BRP4G-cuda32-nv301). BOINC shows me that the estimated time to complete is 00:01:11. However, after 00:23:54 run time (approx. 9.240% completed) BOINC kills the task with the message "Aborting task p2030.20131124.G176.16-01.04.S.b4s0g0.00000_464_0: exceeded elapsed time limit 1432.71 (5600000.00G/3908.68G)" One example of these is result #1454983.
Recently I got a bunch of BRP v1.33 tasks (BRP4G-cuda32-nv301). BOINC shows me that the estimated time to complete is 00:01:11. However, after 00:23:54 run time (approx. 9.240% completed) BOINC kills the task with the message "Aborting task p2030.20131124.G176.16-01.04.S.b4s0g0.00000_464_0: exceeded elapsed time limit 1432.71 (5600000.00G/3908.68G)" One example of these is result #1454983.
There were some problems this morning with the updated server code, and - as Bernd says in message 112866
Quote:
Our plan class specs that were (semi-)automatically converted for the new server code were somewhat broken, causing probably all kinds of oddities for GPU tasks.
This is probably one of them...
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
Recently I got a bunch of BRP v1.33 tasks (BRP4G-cuda32-nv301). BOINC shows me that the estimated time to complete is 00:01:11. However, after 00:23:54 run time (approx. 9.240% completed) BOINC kills the task with the message "Aborting task p2030.20131124.G176.16-01.04.S.b4s0g0.00000_464_0: exceeded elapsed time limit 1432.71 (5600000.00G/3908.68G)" One example of these is result #1454983.
There were some problems this morning with the updated server code, and - as Bernd says in message 112866
Quote:
Our plan class specs that were (semi-)automatically converted for the new server code were somewhat broken, causing probably all kinds of oddities for GPU tasks.
This is probably one of them...
I just downloaded 25 BRP4G tasks with an estimated completion time of 15 seconds, wish it was true =)
I found this in client_state.xml for each of these tasks:
[pre] 280000000000000.000000
5600000000000000.000000[/pre]
If I'm understanding this right the tasks will error out with "exceeded elapsed time limit" when the tasks have run for 20x what they where estimated to take.
I've edited the client_state.xml and added a few zeros to the rsc_fpops_bound value and hope that will give the tasks enough time to actually finish.
Let's see what happens when the host asks for work again.
To follow up on my last post my host has now accumulated over 10 valid BRP4G tasks so the server side estimates have kicked in.
Freshly downloaded BRP4G tasks has an estimated time to completion @ 1h22m12s and the observed completion time is within a few minutes of that when running 2 tasks at a time. So this seems to be working as it should.
Digging a bit deeper the "Average processing rate" is @ 56.766 according to the application details page for host 2267. I didn't take note of the average PFC in the server logs but I believe it was around 3000.
So the server thinks the GPU is about 50 times faster than it actually is?
If one has to guess the speed/power of some component is it not better to assume it's slower than it actually is?
Thanks for reminding me, we need to track down the starting point for the GPU pfc calculations. FWIW CPU starting point is that whetstone benchmark it does.
Queen of Aliasses, wielder of the SETI rolling pin, Mistress of the red shoes, Guardian of the orange tree, Slayer of very small dragons.
(at least one of) the daemons running on the server.
Nothing we can do: I've emailed Bernd.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
Mind you, the explanations
)
Mind you, the explanations offered by both Claggy and Holmis only apply to Gravity Wave (CasA) tasks - they're the only ones which use the locality scheduler.
The workunits reported were - properly for this thread - BRP jobs, which download fresh data with every task. I suspect the real explanation was more prosaic - replacement tasks are put at the back of the queue, and with things being quiet here until testing resumed this morning, there were probably very few active computers plodding their way through that queue.
Anyway, all WUs have been completed and validated now.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
http://www.boincsynergy.com/images/stats/comb-3475.jpg
Recently I got a bunch of BRP
)
Recently I got a bunch of BRP v1.33 tasks (BRP4G-cuda32-nv301). BOINC shows me that the estimated time to complete is 00:01:11. However, after 00:23:54 run time (approx. 9.240% completed) BOINC kills the task with the message "Aborting task p2030.20131124.G176.16-01.04.S.b4s0g0.00000_464_0: exceeded elapsed time limit 1432.71 (5600000.00G/3908.68G)" One example of these is result #1454983.
RE: Recently I got a bunch
)
There were some problems this morning with the updated server code, and - as Bernd says in message 112866
This is probably one of them...
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
http://www.boincsynergy.com/images/stats/comb-3475.jpg
RE: RE: Recently I got a
)
I just downloaded 25 BRP4G tasks with an estimated completion time of 15 seconds, wish it was true =)
I found this in client_state.xml for each of these tasks:
[pre] 280000000000000.000000
5600000000000000.000000[/pre]
If I'm understanding this right the tasks will error out with "exceeded elapsed time limit" when the tasks have run for 20x what they where estimated to take.
I've edited the client_state.xml and added a few zeros to the rsc_fpops_bound value and hope that will give the tasks enough time to actually finish.
Let's see what happens when the host asks for work again.
To follow up on my last post
)
To follow up on my last post my host has now accumulated over 10 valid BRP4G tasks so the server side estimates have kicked in.
Freshly downloaded BRP4G tasks has an estimated time to completion @ 1h22m12s and the observed completion time is within a few minutes of that when running 2 tasks at a time. So this seems to be working as it should.
Digging a bit deeper the "Average processing rate" is @ 56.766 according to the application details page for host 2267. I didn't take note of the average PFC in the server logs but I believe it was around 3000.
So the server thinks the GPU is about 50 times faster than it actually is?
If one has to guess the speed/power of some component is it not better to assume it's slower than it actually is?
Thanks for reminding me, we
)
Thanks for reminding me, we need to track down the starting point for the GPU pfc calculations. FWIW CPU starting point is that whetstone benchmark it does.
Queen of Aliasses, wielder of the SETI rolling pin, Mistress of the red shoes, Guardian of the orange tree, Slayer of very small dragons.
6/6/2014 9:41:16 AM |
)
6/6/2014 9:41:16 AM | Albert@Home | Server error: recompile needed
One CUDA WU errored out after 14h.
What does that mean?
What needs recompiling?
RE: What needs
)
(at least one of) the daemons running on the server.
Nothing we can do: I've emailed Bernd.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
http://www.boincsynergy.com/images/stats/comb-3475.jpg
Server is working again.
)
Server is working again.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
http://www.boincsynergy.com/images/stats/comb-3475.jpg