Hyperscale Infrastructure Q&A
summary
Live Q&A details operational fixes and numbers: dynamic configuration and kill switches plus a company-wide push for push rule that keeps binaries under 60 days old to reduce library deployment pain; capacity planning uses QPS and selective CPU-instruction profiling, a solver with a residual variable and post-solver capacity estimator overrides; service router now handles memcache and Tao after a performance overhaul; core is C++ with sidecar wrappers; sharding via Shard manager; cross-region picks update every 60s and clients use power-of-two choice.
AI-written summary · about
chapters (10)
- 0:00Q&A
- 0:25Service Router deployment challenges
- 2:27Flux capacity planning
- 4:33Why Solar wasn’t UDP
- 6:01Service Router data plane flavors
- 8:26Load testing for capacity
- 9:08Service Router DB/cache support
- 11:18Load balancing compute power
- 12:41RDMA pros and cons
- 13:03Cross-region load balancing
related talks
19mKeynote: Operating 600+ Mesos Servers on 7 Datacenters @Criteo
16mPreview: Cloud Computing
25mTackling operational and capacity utilization challenges at a large scale enterprise deployment
20mManage Scaling, While Keeping Costs Under Control. Challenge Accepted.
25mKeynote: Infrastructure - Introduction
32mBuild without limits with AWS Cloud infrastructure
18mKeynote: Mind the Gap: Bridging Cloud Native Innovation with Real-World Use
33mEnhancing Ceph for Infinite Scalability at LINE