Reduce database size/load, so caches can be increased.


log in

Advanced search

Questions and Answers : Wish list : Reduce database size/load, so caches can be increased.

Author Message
Grant (SSSF)
Send message
Joined: 19 Aug 99
Posts: 5683
Credit: 56,047,915
RAC: 49,958
Australia
Message 1395377 - Posted: 26 Jul 2013, 23:02:44 UTC
Last modified: 26 Jul 2013, 23:03:17 UTC

My understanding is that the serverside limits (100 WU per CPU & 100 WU per GPU)were introduced due to issues with the size of the database. The increased run times of v7 have reduced the number of WUs required to fill a cache, but even so faster systems can still run out of work during the weekly outages, let alone when there are system issues.

While the longer runtimes have reduced the number of WUs pending at any given time, it also means it takes much longer for WUs to be retired from the active list.
I've noticed with several of my pendings the deadline is 2 months from the date of issue. If the WU keeps being given to systems that aren't returning work (for whatever reason) it will be 8 months before that WU times out. Just being given to 2 other non-returning systems before going to one that does return valid work means at least 4 months.


It may be worth the effort to allow the Scheduler to give WU re-issues priority (or only) to systems that have a low average turnaround time and a high number of consecutive valid tasks and/or max tasks per day number.
This should allow WUs to be validated sooner, reducing the number of WUs in the database & the length of time they are there.
And hopefully allowing larger caches once again.
____________
Grant
Darwin NT.

Questions and Answers : Wish list : Reduce database size/load, so caches can be increased.

Copyright © 2014 University of California