We decided to give it a try, not knowing how much we would gain, but we didn't expect such a drawback. And so far none of us understands the bad validation rate. I suspect that the 0.5Hz shift triggers an inconsistency in the app, but may be completely wrong.
Workunits generated from now on will be processed as before (one continuous 1Hz application run). These workunits (and tasks) will again be named "O3ASBu" instead of the previous "O3ASBuB".
Workunits generated from now on will be processed as before (one continuous 1Hz application run). These workunits (and tasks) will again be named "O3ASBu" instead of the previous "O3ASBuB".
These new O3ASBu tasks will be run with the new opencl-nvidia-3 application. Was that the intention?
[edit] The older resends are run with opencl-nvidia-2.
We decided to give it a try, not knowing how much we would gain, but we didn't expect such a drawback. And so far none of us understands the bad validation rate. I suspect that the 0.5Hz shift triggers an inconsistency in the app, but may be completely wrong.
Workunits generated from now on will be processed as before (one continuous 1Hz application run). These workunits (and tasks) will again be named "O3ASBu" instead of the previous "O3ASBuB".
(That's ok.) The *-3 app versions can run both.
However, good point, disabling the older versions could save the costly workunit id check.
I don't know if this is of any consequence or not, but when I was checking my O3AS tasks in the 'Pending' column, some pages were out of sync with the others; i.e. - the count per page were off by a little or quite a bit, as in this page:
I rechecked it several times, and even switched computers, and it was always screwed up the same way. Usually I find that when looking at all the columns about an individual computer, they run 20 tasks per page. But not this time with the 'BuB' tasks.
Just thought I'd throw that out to you if it means anything...
I think we found the problem with validation of the BuB results. Please compute the remaining results. They should be credited correctly. I'll also run a re-validation of the results that were found invalid, these should also be credited correctly.
I think we found the problem with validation of the BuB results. Please compute the remaining results. They should be credited correctly. I'll also run a re-validation of the results that were found invalid, these should also be credited correctly.
oh nice. so it was something with the validator and not the tasks themselves?
We decided to roll back the
)
We decided to roll back the split.
We decided to give it a try, not knowing how much we would gain, but we didn't expect such a drawback. And so far none of us understands the bad validation rate. I suspect that the 0.5Hz shift triggers an inconsistency in the app, but may be completely wrong.
BM
Workunits generated from now
)
Workunits generated from now on will be processed as before (one continuous 1Hz application run). These workunits (and tasks) will again be named "O3ASBu" instead of the previous "O3ASBuB".
BM
Bernd Machenschalk
)
These new O3ASBu tasks will be run with the new opencl-nvidia-3 application. Was that the intention?
[edit] The older resends are run with opencl-nvidia-2.
That's ok. The *-3 app
)
That's ok. The *-3 app versions can run both.
However, good point, disabling the older versions could save the costly workunit id check.
BM
Bernd Machenschalk
)
What would you like to happen with the outstanding BuB tasks?
I assume they should be completed rather than aborted? If you don't make that clear, people may start dumping them anyway.
Cheers,
Gary.
I personally am letting them
)
I personally am letting them run out vs. aborting them.
Phil
I thought I was wrong once, but I was mistaken.
Thank you for doing that.
)
Thank you for doing that. That's the best option until informed otherwise.
Cheers,
Gary.
Bernd Machenschalk wrote:
)
I don't know if this is of any consequence or not, but when I was checking my O3AS tasks in the 'Pending' column, some pages were out of sync with the others; i.e. - the count per page were off by a little or quite a bit, as in this page:
I rechecked it several times, and even switched computers, and it was always screwed up the same way. Usually I find that when looking at all the columns about an individual computer, they run 20 tasks per page. But not this time with the 'BuB' tasks.
Just thought I'd throw that out to you if it means anything...
Proud member of the Old Farts Association
I think we found the problem
)
I think we found the problem with validation of the BuB results. Please compute the remaining results. They should be credited correctly. I'll also run a re-validation of the results that were found invalid, these should also be credited correctly.
BM
Bernd Machenschalk wrote: I
)
oh nice. so it was something with the validator and not the tasks themselves?
_________________________________________________________________________