[
https://issues.apache.org/jira/browse/QPID-7038?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15163871#comment-15163871
]
Keith Wall edited comment on QPID-7038 at 2/24/16 9:56 PM:
-----------------------------------------------------------
The patch looks like a useful improvement to me. On my machine it saves about
20%-30% off the number of iterations to achieve the same answer. I did notice
towards the end the run the algorithm tends test a number of fractional target
rates. Perhaps these should be pruned by have the target rounded to an integer.
{noformat}
016-02-22 22:53:01,824 INFO [org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 16, run number : 3/3, rate: 10517.0
2016-02-22 22:53:22,542 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 17, run number : 1/3, rate: 10515.0
2016-02-22 22:53:43,760 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 17, run number : 2/3, rate: 10515.0
2016-02-22 22:54:04,979 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 17, run number : 3/3, rate: 10515.0
2016-02-22 22:54:26,208 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 18, run number : 1/3, rate: 10514.0
2016-02-22 22:54:47,514 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 18, run number : 2/3, rate: 10514.0
2016-02-22 22:55:08,724 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 18, run number : 3/3, rate: 10514.0
2016-02-22 22:55:29,992 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 19, run number : 1/3, rate: 10513.5
2016-02-22 22:55:51,200 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 19, run number : 2/3, rate: 10513.5
2016-02-22 22:56:12,426 INFO
[org.apache.qpid.disttest.ControllerRunner.main()]
o.a.q.d.c.HillClimbingTestRunner Running test :
TestInstance[testName=persistent_transaction_plain, iterationNumber=0],
iteration : 19, run number : 3/3, rate: 10513.5
2016-02-22 22:56:33,703 INFO
[org.apache.qpid.disttest.ControllerRunner.main()] o.a.q.d.c.Controller
Finished test TestInstance[testName=persistent_transaction_plain,
iterationNumber=0], result TestResult{_name='persistent_transaction_plain',
_hasErrors=false, _producedMessageRate=10070.744160664917,
_consumedMessageRate=10055.246132888324}
{noformat}
was (Author: k-wall):
The patch looks like a useful improvement to me. On my machine it saves about
20%-30% off the number of iterations to achieve the same answer. I did notice
towards the end the run the algorithm tends test a number of fractional target
rates. Perhaps these should be pruned by have the target rounded to an integer.
> [Java Perftests] Speedup Performance test execution
> ---------------------------------------------------
>
> Key: QPID-7038
> URL: https://issues.apache.org/jira/browse/QPID-7038
> Project: Qpid
> Issue Type: Improvement
> Components: Java Performance Tests
> Reporter: Lorenz Quack
> Attachments:
> 0001-QPID-7038-Java-Perftests-Add-Bias-to-HillClimber.patch
>
>
> Currently, running the performance tests takes a considerable amount of time.
> There are some ideas floating around about how we could cut down on the
> execution time. Each of these (and any additional) ideas should be evaluated
> for their impact and implemented accordingly.
> In no particular order; enumeration provided solely for easier reference:
> # because we retry a failed target rate is expensive we might want to change
> the hillclimber to be biased and not divide the range 50:50 but perhaps 25:75
> to decrease the times we miss a target
> # we might want to increase the {{hillClimberMinimumDelta}}
> # we could experiment with using the achieved rate as a guide for the target
> rate for the next iteration
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]