Patrick_Higgins_6715

Performance for Oracle DB Agents

Discussion created by Patrick_Higgins_6715 on Feb 10, 2017
Does anyone have performance data on what can be achieved with a DB Agent?
--- we run over 250k DB Oracle jobs / day using 10 DB agents.   Automic is unwilling to share any details on what was achieved in product testing.   We are currently using 10.0.8 using Oracle 11.2 as the repository

Most of the load is against a 4 node Oracle 12 RAC system hosting a multi-terrabyte "data warehouse".   Three of the agents are dedicated to this database and run 50-60k Oracle jobs / day each.   Also more than a couple of SQL*Loader jobs on each, but I am focusing on the SQL jobs.

The load is mostly queries with updates occurring a couple of times/day.   This DB is moving to an Oracle Exadata configuration and the DBAs have determined that Automic Agents cannot reside on the DB servers (as is the case today).  

Benchmark performance information would help us determine if the 'correct' number of DB Agents for this application is 1 (save on license cost), 3 as it is today but on remote servers, or ? to allow for additional overhead for network connection time between the Agents and the managed database.   Yes we will probably use trial and error (increase/reduce agent count and track the changes in performance) but  would like to have a target number

Eg:  
"Using AE Vxx with a remote Oracle DB Vyy with 1 Oracle agent Vzz on specified hardware.  
Testing reached a rate of xxxx jobs per hour,   with sample end to end job time within 125% of average time for the same jobs at minimal load.   Testing did (or did not) continue till performance degraded"

We have had different issues affecting Automic performance in the past.  Some were issues inside UC4,  more were caused by external 'opportunities' - odd network routing, competing use on database servers, competing use on db storage arrays, bad java versions (the list goes on)

I understand that we will never have the same environment as used in a benchmark, but if our components (server/network/etc) match or exceed the specifications used in the test, I would expect our performance to also match or exceed the results.   If we do not , then we PROBABLY have non-Automic issues that need to be addressed.


Any and all comments/suggestions/detailed test results welcome.

Patrick Higgins


Outcomes