Hi,
I just came across one of the message in elog mentioning that on Sept 2nd,
there was a lrd process on bud20 slowing down online crunching.
I could not remember exactly if that was my process but I have been quite
good at submitting my jobs to pretty much all the buds. Most of the time,
I nice my jobs on bud01-13 and bud20-24 but sometimes I forgot, sometimes
I choose to be greedy a little bit.
My apologize if I came into the way of online crunching.
However, may I also propose that bud20-24 be used only as backup CPUs for
online crunching, that they be only used when cruncher has died for a long
period of time and we have serious backlog. In this case shift crew can
change the 0's to 1's in spuds_list.txt and they be changed back to 0 as
soon as all the runs in the waiting list are sent to the CPUs. My
impression is the 16 spud nodes are well sufficient for normal data rate.
I would like to be able to use bud20-24 with not too low priority, since I
frequently need to recrunch whole month worth of data for my calibration
program. though I have modified my program to reduce the CPU demand by a
LOT, there is still a not insignificant crunching time. BTW, I think it is
well worth it as I started to see improved momentum resolutions.
Thanks.
Chi
This archive was generated by hypermail 2.1.2 : Mon Feb 24 2014 - 14:07:31 EST