Real world PeopleSoft tuning (PSAPPSRV) experience.
In the interest of sharing, I am posting my take on optimizing our PeopleSoft system. I am looking for comments, questions, and, most importantly feedback, from others who have information to add. Perhaps what I have to say is universally understood, I don't know, but if I had seen this information a year ago, I would have had better insight on how to start.
As a relatively new admin to PeopleSoft, it took me some time to realize that the workhorse of our PeopleSoft system is the PSAPPSRV process. I spent considerable time looking for guidelines on optimizing the PSAPPSRV processes, and what I discovered is that there are few on-line guides that really spell out a good approach to sizing the settings for the PSAPPSRV process. The Oracle Red Paper: "PeopleTools Performance Guidelines" (April 2013) gives a good description of how PS system components interact, and has very good advice. I disagree with some of the recommendations for PSAPPSRV management a bit though.
If anyone has references or links to something they found useful, please chime in.
I have been testing and monitoring a production PeopleSoft system for a while now, and I am going to throw out some of what I surmise:
#1 In general, the more PSAPPSRV processes you can launch, the faster your users will get results. Caveat: You must determine the correct number for your system.
#2 Unless you are running multiple different applications on a server (which is probably not a good idea) set MIN/MAX counts to be the same for the PSAPPSRV and all other processes.
#3 Memory and CPU utilization are the simplest measures to use to see if your PSAPPSRV processes are taxing system resources:
#3a If you are regularly running at 100% CPU, you need to scale back: reduce the number of PSAPPSRV processes.
#3b If you exceed physical memory, you need to either scale back, or reduce the recycle count.
#4 Restarting servers and/or clearing cache are an absolute last resort to solving any performance issues. This is treating a symptom, not a root cause, and leads to poor performance for all.
I realize that my findings are a result of the systems I have to work with, and the load that we experience. The above findings, however, held true through a recent conversion from physical, Sun Solaris systems to virtual, Red Hat Linux systems unchanged.
My current (ancient) system:
Tools Release 8.49.10
Application Release HRMS & Campus Solutions 8.90.00.034
Systems supporting these:
Separate App & web servers are VMware virtual systems with 3 web server and 4 app server CPUs and 8G memory. 7 web servers 10 application servers.
Database servers are two 12-CPU blades running Oracle 11g RAC - these are very fast, are well managed, and rarely, if ever, a bottleneck.
The above described application servers run with 20 PSAPPSRV processes and two tuxedo queues per system for 10 PSAPPSRV processes per tuxedo queue.
We use Apache's web server as a front end to the PS Weblogic instances in support of our system-wide Single Sign On solution - a kerberos-based authorization.
We have not spent significant time optimizing the weblogic JVM - currently JRockit 1.5 64-bit.
I think that for much of the *huge* amount of information available on-line regarding optimizing and troubleshooting performance issues for PeopleSoft is difficult to know what vintage system it applies to. There are guides that refer to optimization for systems going back many years - during which time many computer systems and networks have seen 10-times or greater performance improvements. All resources must be reviewed with some skepticism.