07-27-2016 03:13 PM - edited 07-27-2016 03:14 PM
We are struglling to improve performance for our production job, please help.
We started running many more jobs into produciton than before and few jobs which were used to run for ~ 2 hours started running for 6 hours and somtime even for 8 or 9 hours. I looked log in details and found that especially few SAS step like PROC EXPAND taking much more time to get complete than before.
Below is the resource difference where PROC EXPAND step has difference in reource used for same job.
test_a = when job was running for ~2 hours and we had few produciton jobs
test_b = when we added more jobs into production and job started running longer
test_c = when we tested job on different server when no other job running
I found PROC EXPAND step using same resources except "real time", "page faults" and "page reclaims".
Any work around?
|user cpu time (seconds)||44.42||48.91||45.04|
|system cpu time (seconds)||48.97||52.97||36.92|
|OS Memory (k)||14196.00k||14196.00k||14196.00k|
|Voluntary Context Switches||327905||330321||327988|
|Involuntary Context Switches||25059||14045||5942|
07-27-2016 03:45 PM
You need to do more investigation to find the cause or causes for the jobs' slowness. You need to monitor the performance of your SAS Application server while these jobs are running. The best tool to use for this is SAS Environment Manager. Other tools can also be useful for example Windows Task Manager, Perfmon and so on.
The most likely causes for your slowness are either IO limitations or CPU usage reaching close to 100%. Monitoring the server performance will allow you to identify the bottlenecks.
07-29-2016 12:53 PM
From the stats you sent in, your computer system is bottlenecked on a computer resource (IO, CPU or memory). You need to monitor your computer system with a hardware monitor tool (like IBM's nmon or Microsoft's PerfMon). Papers on how to use these tools can be found on this SAS Usage Note http://support.sas.com/kb/53/877.html
Let us know if you want help reviewing the monitor data.