XNAT Tuning & Monitoring
- 1. XNAT Tuning & Monitoring John Paulett [email protected] June 29, 2010
- 2. Overview Share NRG’s experiences running a large XNAT installation, including methods for tuning, testing, and monitoring the application.
- 3. Plan 1. Sample XNAT Architecture 2. Hardware “Recommendations” 3. Monitoring XNAT 4. Performance Testing Tools 5. Tuning XNAT
- 4. XNAT SCALES!
- 5. XNAT Scales From a single study with dozens of scans To hundreds of studies, including large, multi- site studies
- 6. Central Neuroimaging Data Archive (CNDA) Flagship XNAT installation at Washington University (WUSTL) As of June 2010: – 500 studies – 8000 subjects – 11k imaging sessions – 9 TB of data
- 8. CNDA Architecture • 1x Kemp load balancer – SSL acceleration • 1x Quad-core Xeon, 16GB RAM: PostgreSQL 8.3 • 2x Dual-core Xeon, 4GB RAM (one in standby): Tomcat 5.5 & DicomServer • BlueArc NAS • Multiple Sun Grid Engine nodes
- 10. Future Architecture Reduce single points of failure – Standby Kemp load balancer – PostgreSQL Warm Standby – Actively load balance Tomcat – Archival storage SAN Use new super computer at WUSTL Improve ability to upgrade without downtime
- 13. Grow into your Architecture Get single good server – Multicore with 4-16GB RAM (better than central.xnat.org) – Consider your archive’s future size & location When you outgrow: – Buy a more powerful machine for PostgreSQL – Leave Tomcat on first server
- 14. MONITORING XNAT
- 16. Pingdom • World-wide tests for site availability & response time • SMS & Email alerts when sites are unavailable
- 17. Munin PostgreSQL, Tomcat, & Linux metrics over time – Memory, CPU, queries, requests, etc.
- 20. Monit Active process monitoring & management Define criteria for emailing alerts & restarting processes – CPU, memory thresholds – Connection failures (check web services)
- 22. JMeter Generate load & analyze throughput Complex HTTP transactions
- 23. JMeter https://svn.apache.org/repos/asf/jakarta/jmeter/
- 24. YourKit Profiling Lower level debugging tool Memory & CPU profiling Hunt down memory leaks & code hot spots Can instrument in production server
- 26. TUNING XNAT
- 27. On Tuning Tuning results dependent on many variables, what worked in one case may not work universally XNAT is a complex system! Some parts are CPU bound, others are memory bound, and others are bandwidth bound General rule: faster CPUs + more RAM + bigger network pipe = faster XNAT
- 28. On Tuning 1. Find something that is “slow” 2. Quantify slowness 3. Tune 4. Quantify improvement 5. Go to #1
- 29. PostgreSQL Tuning PostgreSQL 8.3 has serious performance improvements Put PostgreSQL and Tomcat on separate machines – Get powerful database machine – PostgreSQL can take advantage of multiple cores & lots of RAM
- 30. postgresql.conf Default settings designed for legacy machines Increase available memory. Allows query planner to do more work in RAM and less on disk Increase max connections Tweak kernel settings to allow access to more memory
- 31. Tomcat Tuning Increase available memory & use “server VM” catalina.sh: JAVA_OPTS="$JAVA_OPTS "-XX:MaxPermSize=256m" "- XX:PermSize=256m" "-mx1512m" "-server"
- 32. Tomcat Tuning Increase connections & threads in server.xml – At this point, consider load balancing between multiple Tomcat servers
- 33. XNAT Tuning Upgrade to XNAT 1.4! Increase MaxConnections to database in WEB-INF/conf/InstanceSettings.xml – Set in line with PostgreSQL’s max_connections
- 34. Tools Mentioned Google Analytics: Free Pingdom: Monthly subscription. One site free Munin: Open Source Monit: Open Source JMeter: Open Source YourKit: Commercial. Time-limited trial & free for open source
- 35. Questions & Your Experiences? http://www.xnat.org/XNAT+2010+Workshop+- +Tuning,+Optimization,+Monitoring