1) Message boards : Number crunching : extreme long wu's (Message 2501)
Posted 20 Nov 2017 by xaminmo
I wish I'd looked more deeply into this before joining up.

My favorite is a task listed as 30 days estimated runtime, but a due-date 14 days into the future.

Of course I aborted that, but I shouldn't have to micromanage compute. They should be well behaved, self policing, etc.

I did bump up to 7.8.4 client, just in case (kind of a hassle on xenial, but not too bad).

It looks like U@H is not the only problematic project, but seems to have the biggest problems.

IIRC, UID 1 has backed off of the project due to family constraints, and UID 2 basically owns the project.

My guess is they just really do not have the time or resources to troubleshoot this fully, but are hesitant to kill the project either.
2) Message boards : Cafe : 29 day task due in 12 days? (Message 2500)
Posted 20 Nov 2017 by xaminmo
More digging, and this seems to be a known issue.

SOP seems to be to abort any WU with an estimated runtime over 6 or 8 hours.

All of the frustrations have already been discussed, with the most exhaustive thread being here:

It seems either monitor closely all the time, accept wasted cycles, or set to "No new tasks".
3) Message boards : Cafe : 29 day task due in 12 days? (Message 2499)
Posted 20 Nov 2017 by xaminmo
Anyone else getting impossible runtimes? I have jobs running from 2-4 days, which seem kind of greedy, but then I have a task that says 19 hours elapsed, 28 days and 11 hours remaining, due on December 2.

It's not even worth calculating, because it's impossible to complete.

I have 24 cores, and all of the U@H work items are single CPU.

I'm all NVIDIA, so the ATI jobs don't run for me.


Copyright © 2024 Copernicus Astronomical Centre of the Polish Academy of Sciences
Project server and website managed by Krzysztof 'krzyszp' Piszczek