Since upgrading a reference node AIX from version 5.6 to version 6.10.61, the customer has noticed a drift in the required time to end a job during their nightly batch, starting around 21:00. For example, a job whose script has ended at 21:04 will have the completion instruction complete at 21:16.
Since this is AIX, many tuning instructions were applied but the problem reocurred at night ( during production peak).
These are the variables added for tuning reasons:
Customer has noticed that UVC is speedier, but the time drift remains. This impacts the majority of jobs after 21:00, though some wait longer times between script completion and completion instructions than others.
We don't see any uxjobend processes that are dragging on, what could be the cause?
Faulty U_POST_UPROC, this could be seen by adding some date commands in the U_POST_UPROC and checking the Job Log, to check how long it would take each command to complete.
Release : 6.x
Component :Dollar Universe
The issue due to the U_POST_UPROC that was far from optimal.
On this case, customer was launching 6 Dollar Universe commands (5 uxlst ctl + 1 uxlst upr) for each Uproc execution to fill a file with details about the ongoing execution, this generated an additional overhead to the IO server for nothing during production peak (around 21:00) for nothing.
Instead, a echo in the u_ante_uproc and u_post_uproc regarding the current uproc and timestamps for start and end could be used for monitoring in a txt file if necessary, else use Reporter which does this monitoring natively.
Else, simplify the U_POST_UPROC script to reduce the 5 uxlst ctl commands to only one or rename the U_POST_UPROC.