Systems Scale Summer Q&A Week 1
summary
Q&A summarizes concrete practices: Netflix open‑sources Netflix concurrency limiters and uses a runtime monitor that auto-tunes per‑host limits; load balancing uses a simple pick-two algorithm counting outstanding requests to pick the least loaded peer; AWS Lambda enforces a Lambda max runtime and recommends clients send timeout headers so functions stop when clients give up; prefer quick rejection over queuing (Little’s law explains why queues bloat), and production load testing splits responsibilities between a load controller (desired rate) and an actuator (traffic weights).
AI-written summary · about
chapters (23)
- 0:00Session introduction
- 0:26Service level objectives with user journeys
- 1:30Dynamic concurrency limits and Netflix tools
- 2:36Load testing beyond capacity measurement
- 3:43Circuit breakers and overload protection at Amazon
- 5:27Wasted work in overloaded systems
- 7:32Lambda architecture and load balancer usage
- 8:14Pick two load balancing signals
- 9:18Tenant fairness in multi-tenant systems
- 11:23SLOs for services with multiple dependencies
- 12:44Preventing load test production disruptions
- 14:29Reproducibility with live traffic load testing
- 15:34Lambda timeouts and client detection
- 17:02Adaptive load balancing for spiky workloads
- 19:48Actuators and traffic control interaction
- 21:34Load controller vs actuator separation
- 23:02User statistics for SLO determination
- 24:47Rejecting excess traffic vs queueing requests
- 27:16Validating dynamic concurrency limits
- 28:40Lambda resource constraints handling
- 30:25Sudden production load surges
- 32:10Retries impact on reliability and overload
- 34:19Closing remarks and week two preview
related talks
21mHandling Traffic Spikes Through Automated Prioritized Load Shedding
44mTechniques Netflix Uses to Weather Significant Demand Shifts
32mSystems @Scale Summer Live Q&A Session #4
39mPractical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix
21mSvalinn: Overload Control in Large-Scale Servers with Multiple Resource Bottlenecks
26mEfficient load balancing in distributed systems
40mEverything Will Flow
01h 16mInterview and Q&A