The Show Must Go On: Load and Optimization of an Online Conference
summary
A team from JUG Ru Group built an online conference platform used for JPoint and Joker events and for external meetups, with a hard requirement to support 10,000 concurrent attendees with high availability. The talk describes the platform’s main components, why Gatling plus NanoCloud were chosen for load testing, how the load profile was constructed, and what component- and infrastructure-level optimizations were implemented to approach the 10K concurrent-user target.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 2:31Speaker and context
- 4:16Life 4 platform evolution
- 5:18Load testing goals
- 6:00Online broadcast architecture
- 10:04Load testing tool selection
- 12:33Load testing scenario
- 14:16Bot preparation and launch
- 17:42NAN Cloud load distribution
- 24:30First results and issues
- 25:33Tomcat and WebSocket optimization
- 28:24Internal traffic problems
- 30:29MongoDB data aggregation load
- 36:57HeelCast async insert issues
- 38:45HC for persistence
- 39:47Kubernetes network call optimization
- 41:30Redis master-replica split
- 43:15HTTP client connection tuning
- 44:41Summary and conclusions
- 46:49Future plans
related talks
21mHow Do We Run Online Conferences?
10mWrapping Up the Online Part of JPoint 2024
36mToPlay on Google Cloud
42mTuning and Optimizing High-Load J2EE Web Applications
22mBuilding a Flexible, Scalable Infrastructure to Test Dolby.IO at Scale
42mLoad Testing Is Complicated: A Case Study of NJ Unemployment Insurance
01mInterview
43mGoogle - Load-shedding