Troubleshooting

Improving Oracle OLTP database performance with Dell Fluid Cache for DAS
13
8.1 Before each test, it was ensured that the Open Manage ‘dataeng’ service was
stopped to avoid any performance overhead.
8.2 Before each test, it was ensured that Oracle AWR report was not being collected to
avoid any performance overhead.
9. Test start:
9.1 The OSWatcher script was started on the database server to capture the performance
metrics.
9.2 The OLTP workload setup in step 8 was started from the BMF server.
10. Test end: The test was allowed to run until the average response time (ART) as displayed by
BMF crossed the 2 seconds limit.
11. Post-test results capture:
11.1 The User load, TPS and the Average Response Time (ART) results were captured using
BMF.
11.2 CPU utilization, IOWaits, and memory utilization stats were extracted from OSWatcher
results using a custom script.
12. Re-running the test:
12.1 Database was restored to the clean state using the flashback guaranteed restore point
created in step 6.2
12.2 Server was rebooted to ensure the server is in a clean state
12.3 Repeated step 8, step 9, step 10 and step 11 in that order to test the solution again, if
necessary.
Fluid Cache based solution test methodology
This section describes the test methodology that is specific to this solution testing. Hardware,
software, test methods and configuration settings listed in the Baseline configuration test methodology
section apply to this solution testing as well, unless specified below.
1. Restore database: Database was restored to the clean state before installing or enabling Fluid
Cache for DAS caching using the flashback guaranteed restore point created in step 6.2 of the
Baseline configuration test methodology section. This was done to ensure cold cache.
2. Express Flash PCIe SSD hardware and software setup: Only two 350GB Express Flash PCIe
SSDs were used for this solution testing giving a cache pool size of ~650GB. Fluid Cache
software was installed using the Fluid Cache for DAS user’s guide
8
.
3. Fluid Cache setup: Fluid Cache was tested in write-back mode only. For more details on the
configuration steps, refer to Appendix C.1.1 Enabling Fluid Cache.
4. IO Queue Depth: The Fluid Cache IO queue depth was increased from the default value of 64
to 256 in the /sys/kernel/config/fldc/io_queue_depth parameter file.
5. Testing and results capture: Step 12 from the Baseline configuration test methodology section
was repeated to test this storage solution. This step corresponds to restoring the database to a
clean state, restarting the OLTP benchmark and capturing the results.
6. Re-running the test:
6.1 To flush all the data block references and to start with cold cache again, Fluid Cache
was disabled. For details on the steps followed, refer to Appendix C.1.2 Removing
Fluid Cache.
6.2 Database was restored to the clean state using the flashback guaranteed restore point
created in step 6.2 of the Baseline configuration test methodology section.
6.3 Rebooted the server to ensure the server is in a clean state.