[
https://issues.apache.org/jira/browse/DERBY-5471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13282001#comment-13282001
]
Rick Hillegas commented on DERBY-5471:
--------------------------------------
Thanks for continuing to test the performance of sequences and identity
columns, Mamta.
Note that the default size of the pre-allocation range in 10.9 is 100, not 20.
We had to boost the pre-allocation size in order to compensate for the
throughput drop introduced by DERBY-5494. The pre-allocation size was boosted
by patch derby-5493-02-aa-boostPreallocationTo100.diff, attached to DERBY-5493.
You should re-run your sequence tests on 10.9 (or the 10.10 trunk) with a
pre-allocation size of 100.
Note also that pre-allocation for identity columns was backed out of 10.9 by
patch derby-5687-01-aa-backOutImprovement.diff, attached to DERBY-5687.
Adjusting the size of the pre-allocation range should not affect the throughput
of identity columns. Your results confirm this.
Note also that 5471-performance.html was attached on 2012-04-17, a couple hours
before Mike committed the work on DERBY-5494. The results in
5471-performance.html confirm that identity columns had identical throughput on
10.8.2 and on the trunk BEFORE Mike committed the DERBY-5494 work. Your tests
measured performance degradation of identity columns AFTER Mike's fix, although
we can't say yet how much of that degradation was due to DERBY-5494 and how
much might be attributed to other changes on the trunk since then. In order to
tease out that information, you might want to run your experiments on the trunk
with DERBY-5494 backed out.
Hope this helps,
-Rick
> Stress test for identity columns and sequence seem to be taking longer on
> trunk compared to 10.8.2.2 RC3
> --------------------------------------------------------------------------------------------------------
>
> Key: DERBY-5471
> URL: https://issues.apache.org/jira/browse/DERBY-5471
> Project: Derby
> Issue Type: Task
> Components: Test
> Affects Versions: 10.9.1.0
> Environment: Windows XP version 2.18
> Genuine Intel(R) CPU T2600 dual core @2.16GHz 2.00GB of RAM
> $ java -version
> java version "1.6.0"
> Java(TM) SE Runtime Environment (build pwi3260sr9fp1-20110208_03(SR9 FP1))
> IBM J9 VM (build 2.4, JRE 1.6.0 IBM J9 2.4 Windows XP x86-32
> jvmwi3260sr9-201102
> 03_74623 (JIT enabled, AOT enabled)
> J9VM - 20110203_074623
> JIT - r9_20101028_17488ifx3
> GC - 20101027_AA)
> JCL - 20110203_01
> Reporter: Mamta A. Satoor
> Attachments: 5471-performance.html
>
>
> I have been trying to run org.apache.derbyTesting.perf.clients.Runner (which
> provides ways to stress test sequence generator and identity columns) on
> trunk and 10.8.2.2 RC3 to compare the performance and I find that it takes
> almost double the time for the tests to finish on trunk. Additionally, the
> identity column test consistently ran into lock timeouts on trunk. I am
> running with insane jars on trunk and 10.8.2.2 RC3. The test in question is
> not in official jars for the release candidate so I manually copied them to
> 10.8.2.2 RC3 environment during my test(basically copied the entire
> org.apache.derbyTesting.perf.clients.Runner directory from trunk to 10.8.2.2
> RC3 environment).
> Command to do sequence stress testing is as follows
> java org.apache.derbyTesting.perf.clients.Runner -driver
> org.apache.derby.jdbc.EmbeddedDriver -init -load seq_gen -load_opts
> debugging=1,numberOfGenerators=5,tablesPerGenerator=10,insertsPerTransaction=100,identityTest=0
> -gen b2b -threads 10
> Command to do identity column stress testing is as follows
> time java org.apache.derbyTesting.perf.clients.Runner -driver
> org.apache.derby.jdbc.EmbeddedDriver -init -load seq_gen -load_opts
> debugging=1,numberOfGenerators=5,tablesPerGenerator=10,insertsPerTransaction=100,identityTest=1
> -gen b2b -threads 10
> An average run on 10.8.2.2 RC3 for sequence stress test is about a minute and
> 46 secs. On trunk, on an average it takes 2 and half minutes
> An average run on 10.8.2.2 RC3 for identity stress test is about a minute and
> 50 secs. On trunk, on an average it takes 3minsutes and 30 secs. Also, on
> trunk, this test runs into lock timeouts.
> I was wondering if this is the right behavior. The performance should be
> better in trunk because of pre-allocation of range for sequences and identity
> columns(which defaults to 20) but unless I have missed something in my tests,
> the results don't show the performance improvement.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
