[torquedev] Patch: spread job polling more uniformly
rea+maui at grid.kiae.ru
Sat Jan 21 13:07:31 MST 2012
It was a long time since I had posted to this list, but now I have
a patch that should help a busy scheduling systems to be more
responsive. During the optimization of our Torque/Maui server,
I had found that the spread of the running job polls is governed
by the queue_rank value: the remainder from its division by
JobStatRate is used as the spreading factor.
That's not a very good thing, because we can do better by just
spreading the polls really uniformly:
t_n = now + n * JobStatRate / N_running.
The attached patch does this. It is currently being tested on
our cluster and shows no regressions for some hours. I'll test
it more thorougly, but the code review will also be good.
Eygene Ryabinkin, Russian Research Centre "Kurchatov Institute"
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 3673 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torquedev/attachments/20120122/9bb79821/attachment.bin
More information about the torquedev