Fermi LAT Gamma-ray pulsar search #4 "FGRP4"
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4332
Credit: 251670587
RAC: 35424
18 Aug 2014 13:17:22 UTC
Topic 197685
(moderation:
We will soon begin to test the new Gamma-Ray pulsar search FGRP4. The goal is to have that tested and running without problems by the end of this week.
During Test this will run as "beta" application, i.e. if you agreed to receive "beta / test application versions", you'll probably get FGRP4 work, too.
Until further notice this application will be CPU only, there will be no GPU (i.e. OpenCL) versions.
BM
Edit: I thought that I announced this search somewhere in this thread, but actually can't find it there. Anyway, it's a continuation of the FGRP search, analyzing new, more sensitive data published by the FERMI collaboration recently, with some improved search code.
BM
Fermi LAT Gamma-ray pulsar search #4 "FGRP4"
)
Thanks for the info and I think this is might be the message you're thinking about.
Thanks. The message I was
)
Thanks. The message I was looking for is this. I didn't mention "FGRP4" explicitly there, which is probably why I missed it at first.
BM
BM
I got my first batch of FGRP4
)
I got my first batch of FGRP4 tasks today and curious about how they would run I promoted one of them by suspending other tasks in the queue.
The task fail after about 38 min with a "197 (0xc5) Maximum elapsed time exceeded" error.
The estimated time to completion is 2min31s with a DCF of 1.2991 and the task reached about 67% into the second sky point out of 3.
I have now inoculated the rest of my assigned tasks by adding a couple of zeros to the "rsc_fpops_bound" value in client_state.xml hoping for more success.
The inoculation worked and
)
The inoculation worked and the first task completed at about 1h9min elapsed time and gave a whooping 2.58 credits. Still some tuning to do before this is ready for production release. =)
I got a batch of FGRP4 tasks
)
I got a batch of FGRP4 tasks on this laptop.
My first one ran to completion successfully--need the quorum partner to run to see whether things went right.
Observations:
1. the initial completion time estimate was very far low--something like 6x.
2. the credit awarded (as seen here and elsewhere) of 2.58 seems very low in relation to the CPU work required.
3. I'm very happy to see CPU work available in small enough doses of computation required to be suitable either for lower output machines which run 24/7, or for somewhat higher output machines which run intermittently (as does my laptop).
RE: The inoculation worked
)
I was interested to see that you've finished all the FGRP4 tasks you initially received and they were mainly the 'short ends' type, by the look of things. You actually did have one task in the batch that took about ~8x longer and requested ~8x more credit.
I was also interested to see how many had already been validated. To me that means that other hosts in the main are managing to escape the dreaded 'max time limit exceeded' problem that hit you. Perhaps that's because your host is running with HT enabled and is doing GPU tasks on both the external and internal GPUs which is slowing down your CPU tasks. Your quorum partners are perhaps not being slowed in this fashion and (by good luck) are managing to beat the limit :-).
The estimates for all the other runs on your host must have really got stuffed up by the behaviour of this new FGRP4 run. Thanks for posting your experiences. It convinced me not to even dip my toe in the water just yet :-).
Cheers,
Gary.
RE: RE: The inoculation
)
Nice catch, I missed that one until now.
Yeah, running tasks on both the iGPU and the GPU seems to slow things down a bit even though I've set things up to only run 5 CPU tasks when also doing GPU work. I think there is room for some efficiency improvements on my system but it runs good enough so I haven't bothered trying to squeeze the last percentage out of it.
The DCF landed on about 24 so all estimates got adjusted to about 20 times there normal value. I resorted to a manual edit of the client_state.xml to reset it to 1.5 to get things back to normal.
Thanks for the information on
)
Thanks for the information on the new project.
Since yesterday, my host has completed 36 of the new FGRP4 tasks. 19 of those tasks have validated and the rest are in a pending status.
All version 1.02 FGRP4 tasks
)
All version 1.02 FGRP4 tasks are erroring out almost immediately on my 2 windows/Intel machines.
RE: All version 1.02 FGRP4
)
Something seems to be wrong with the Windows version. Disabled it for now.
BM
BM