11am - 251
Alan Berg (U of Amsterdam)
Linda Place (U of Michigan)
Jean-Francois Leveque (UPMC)
John Fontaine (Blackboard)
Megan May (Indana U)
Ken McPherson (Campus EAI Consortium)
Rob Stonham (University of Hull)
Chris Kretler (U of Michigan)
Participants are encouraged to join the performance working group on collab so that we can communicate together.
Sample load test scripts are created and designed to be run against standard Sakai builds. In this case Sakai 2.5.1 rc1.
- John: Believes we should focus on scenario testing initially, and do tuning later.
- Tuning is done with individual tools with developers involved and iteratively.
- Tool specific testing (w/more granular hardware specific testing)
- hardware parameter tuning tests.
- Jean-Francois: wants provisioning tool to set locale for the users. Is thinking about localization as a performance issue.
- Linda: directions, pseudo-code, properties files, provisioning scripts, load test scripts need to be published.
- We will need to setup a SVN instance.
- Some things will need to be on confluence for edit and easy read (pseudo-code)
- Instructions document on scenario building. (cwk: event counts per transaction?)
- Ken: perhaps some recommended tool combinations (clusters) for test scenarios.
- What do the results look like (how will we publish our results):
- Ken: Suggests TPC model. (http://www.tpc.org/), OLTP
- Dashboard for results sorted by size, tools. Details that can be drilled into.
- Core stats, vs expandable stats.
- Top Ten TPC-C Results by Performance ( All / Clustered / Non-Clustered )
- Top Ten TPC-C Results by Price/Performance ( All / Clustered / Non-Clustered )
- Ten Most Recently Published TPC-C Results
- All Results (by Hardware Vendor/by Database Vendor)
- Advanced Sorting
- Ken: What should we be measuring? items like:
- what to measure? what points to measure? important vs nice to have.
- scripts to support these measurements?
- Diagnose jmx measuresments of class object size.
- where clause on sql statements.
- Some common problems seen in production:
- abandonment - saturate assessment what rate user abandon users stop using it.
- session is getting held, the not in use. session closure.
- all the data is stuck the JVM - garbage collection doesn't happen.
- Alan: Are there sizing guidelines for Sakai?
- lmp: No, people ask large institutions for advice on recommended hardware configuration.
sites: https://confluence.ucdavis.edu/confluence/display/UCDSAKAI/Sakai+Performance+Issues https://confluence.ucdavis.edu/confluence/display/UCDSAKAI/Sakai+Performance+Tuning https://collab.sakaiproject.org/portal email@example.com?subject=Request%20regarding%20collab.sakaiproject.org