I'm still wondering why the ATLAS node that I'm waiting on for my first validation doesn't seem to be using any of its four CPUs for crunching - I see pleanty of valid BRP3cuda32fullCPU results, but all the CPU tasks (which are still being allocated) end up past deadline and self-aborted. Seems a waste of good bandwidth, somehow.
Hm. This is one of the machines we are trying out a new condor configuration for GPUs on. Looks like there's still something wrong with that.
On the Server Status page became allready added a collumn "FRGP1" in the table "Workunits and tasks". But I believe there is still missing a table "FRGP1 search progress" with the absolute, relative and time data?
The list of applications is taken directly from the database, as soon as the application has been set up, it will show up there.
In contrast, the progress counters are set up manually and differently for every search. We are still fiddling with the workunit setup for FGRP1, so we don't know ourselves how many workunits there will be, not to speak about the progress rate.
1) Why is CPU/Work so relative small? Does LAT uses partially also the GPU?
2) Even for CPU-Time the LAT-tasks seems to be underprvileged in earnage. This may be a problem for some poeple.
I would not trust the time statistics for the tasks at the moment because of the bug in the checkpoint *accounting* described above (checkpointing itself seems to be ok). Still this is something to have an eye on, e.g. checking in task manager or "top" on Linux whether a Fermi task is somehow getting less CPU than other apps (because of IO perhaps or whatnot).
Yeah, but don't go use that one as it is the non-checkpointing app that people above pointed out. If you want to go add it to your app_info.xml, wait for a stable version.
My task from wu 100294686 completed and reported. My quorum partner runs an i7 and has little work in queue, but reported several errors on GW work on July 4 and almost nothing on July 5. So confirmation might take a while.
Reported CPU time on my host was 28,896.21. This host is currently typically taking somewhat over 19,000 seconds on GW work.
My quorum partner reported this work at 9 Jul 2011 15:13:59 UTC, and we both validated. Just as an example of the CPU time reporting problem discussed in other posts, his CPU time reported is 1,226.39 seconds!
RE: I'm still wondering why
)
Hm. This is one of the machines we are trying out a new condor configuration for GPUs on. Looks like there's still something wrong with that.
BM
BM
RE: On the Server Status
)
The list of applications is taken directly from the database, as soon as the application has been set up, it will show up there.
In contrast, the progress counters are set up manually and differently for every search. We are still fiddling with the workunit setup for FGRP1, so we don't know ourselves how many workunits there will be, not to speak about the progress rate.
BM
BM
Yesterday my first LAT-file
)
Yesterday my first LAT-file became validated and so does allow some comparison.
* LAT- ---------53715------------30912---------57.55%-------40.76-----------23.46
* S5GC1-------21469------------19653---------91.54%-------45.96-----------42.08
Dimensions: Time [s], earnage [Cobblestone/h]
1) Why is CPU/Work so relative small? Does LAT uses partially also the GPU?
2) Even for CPU-Time the LAT-tasks seems to be underprvileged in earnage. This may be a problem for some poeple.
Kind regards
Martin
Hi! I would not trust the
)
Hi!
I would not trust the time statistics for the tasks at the moment because of the bug in the checkpoint *accounting* described above (checkpointing itself seems to be ok). Still this is something to have an eye on, e.g. checking in task manager or "top" on Linux whether a Fermi task is somehow getting less CPU than other apps (because of IO perhaps or whatnot).
CU
HB
As I use an app_info.xml file
)
As I use an app_info.xml file to run 3 x WU's on each of my GTX 460's...
How can I find this new "APP" to manually download ?
TIA
dunx
RE: How can I find this new
)
See here.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
Yeah, but don't go use that
)
Yeah, but don't go use that one as it is the non-checkpointing app that people above pointed out. If you want to go add it to your app_info.xml, wait for a stable version.
RE: My task from wu
)
My quorum partner reported this work at 9 Jul 2011 15:13:59 UTC, and we both validated. Just as an example of the CPU time reporting problem discussed in other posts, his CPU time reported is 1,226.39 seconds!
RE: RE: How can I find
)
Thank You for the link !
Will try as soon as the coffee starts to work !
dunx
The App 0.17 sent out minutes
)
The App 0.17 sent out minutes ago should have most issues of the 0.16 fixed (checkpointing).
Result's won't validate against these of the 0.16 version, but this should only affect < 5 WUs.
This won't be the 'final' App either, as one important feature is still not working. Anyway, it should be good enough to find the last major issues.
BM
BM