Analytics Performance

ALL installations of Oracle Analytics require some form of performance tuning, either as part of the the initial build, or as a separate exercise.

However, if you have not tuned prior to bringing users onto the system, you may find yourself responding to performance issues, and need to resolve them quickly.

So what do we need to Tune?:

• Operating System
• Database
• Analytics
• Web Servers
• Network

Process

The process of performance management is to Update Settings (aka Tune) then test.  

Tune then test. Tune then test…

Its a long slow process, but there are a few shortcuts, which I will go over in a set of separate posts.
If you already have a live system with active users, then it’s monitor and improve.

Monitor and improve. monitor and improve…

In order to conduct the above process we need to know how to conduct testing and how to analyse the results of the tests.

Testing

The testing part is crucial. Follow the rules

• One change, one test. Don’t be tempted to update the database and do an analytics change at the same time
• Consistent Test. Run tests that are repeatable, in exactly the same way each time it is run. Don’t use randomness in the test.
• Representative Testing. The tests need to reflect the actual use, in the same mix. Don’t just test 10 dashboards with 2 users, if the real pattern is that 20 users use one dashboard, and the other 9 dashboards are rarely used.
• Isolation. Don’t allow outside influences to effect the testing. For example other users of the database having access while you are testing.
• Record the Results! Don’t forget to set up and understand the usage tracking process. Make sure you know what each measure means, and what you are focusing on

The tools you can use:

• URL Based testing. This enables a full test of the whole system, from a users perspective, using methods/tools
  • Load Runner
  • Manual Running
  • curl scripts
  • Apex based
• Component testing
  • nqcmd. this tests the BI Server and database. see Use nqcmd to Test and Refine the Repository
  • Database SQL Running. this is only for SQL tuning

What to test

Testing requires content. Data, lots of data.

And BI Content.

Dashboards, Workbooks, Analyses, Agents, BI Publisher reports.

You choose the order of the things you test and tune, my preference is:

1. Operating System – use CPU and memory stats
2. Database – use DB Time to measure and analyse using OEM
3. BI Server – Use Total Time to measure. Use nqcmd to test
4. Presentation Server – Use Response Times to measure. Load Runner to test
5. HTTP Server
6. repeat the above 1 to 5

Test results

Testing produces results, results that can be compared. These results can be recorded in Usage tracking, in the tooling (e.g. Load Runner results), in script logs, or by simple observation and in recorded stopwatch times.

finally

Whatever you do, you must test in a controlled manner and know the one thing that you’re testing for and whether that one thing improved the system. And remember, performance testing can take a significant amount of time, allow for this in your project plans! Anyway, let’s get stuck in. 

Starting Point

Oracle have produced a tuning guide that should be your starting point. Some of the examples shown below originate from that guide

See it at https://support.oracle.com/epmos/faces/DocumentDisplay?id=2866848.2

There are also some excellent blogs written on OBIEE performance, that are still relevant with OAS (Which is OBIEE 12 except in name!). I suggest you read those blogs too.

---