Design Time - Even though already in Production, Design Time impacts runtime performance. 

• Use TCS steps to break long transactions after sub-process completes, etc.
• Use Delayed Release EAIJAVA call instead of immediate release.
• Use SW_QRETRYCOUNT to handle EAI Java errors.
• Do not delete a step in a procedure. Just delete the incoming link. A lot of times cases have a step outstanding and that step itself gets deleted which causes the instructions to fail. 

Runtime environment - 

• Resource monitoring - Per process / Overall iProcess server / If VM then ESX server as well.
  • CPU
  • Memory
  • Disk I/O
  • Disk space left on device
  • Network latency between servers hosting - iProcess Engine, Database, TIBCO BusinessWorks, TIBCO EMS
  • Time creep between servers
  • Database performance
• Monitor iProcess MBox queues - swadm count_messages ALL
• Monitor WIS process (as suggested in - https://docs.tibco.com/pub/ipe-oracle/11.6.0/doc/html/tib_ipe_admin_guide/wiswqs.16.4.htm#1074729)
• Monitor / Review iProcess logs - sw_error, sw_warn and swentobjsv*logs
• All monitoring data should be compared with an acceptable performance benchmark data at a defined peak load.
• Cleanup file system regularly - Archive old logs / temp files
• Purge cases regularly via dedicated BGs
• Delete all old procedure versions. Archiving should be performed elsewhere.
• Regular database maintenance - update stats / re-org tables / rebuild indexes / monitor internal Oracle processes, etc/
• User administering iProcess should be familiar with at least swadm / swsvrmgr commands (details in iProcess Administrator Guide)
• Change the EAI Java deadlock timeout to 30 secs.
• Version 11.5.0 and higher -
  • Utilize the $SWDIR/etc/sqloptim file to fine tune hints for the available sqlids.  
  • Define a default ONFULL and MAXSIZE attribute for all log files to prevent any from reaching the 2GB limit or just making sure they are rotated/saved appropriately if no value is set.
  • Define max sw_warn/sw_error log file sizes with WARN_ERROR_LOG_SIZE to prevent large warn/error log files. 
• Version 11.4.1 and higher -
  • Monitor the sw_warn file for dequeue times (DB tuning), overall call times (likely due to the need for delayed release), mutex acquisition times (sharing SPO logins).  
  • Use auto-purge deadlines to purge old cases to prevent unneeded cases from piling up in the system (or overall using AUTO_PURGE_DELAY).
  • Use swadm EVLOOPBACK to test events (mostly Oracle related to know to re-start the DB or contact Oracle). 
  • CCOUNT_CACHE_REFRESH now has a value of 0 for everything but RPC_POOL.  For performance, and if this was upgraded, make sure it is not set to a different value for other processes.  
  • If you have a lot of cases, DISABLE_CASE_COUNTS can also be useful (along with disabling case counts in WS(B)).  
• Version 11.4.0 and higher -
  • Set the SPO_CACHE_PROC attribute to prevent the SPO server from caching all versions of every procedure.