Towards Performance Robustness for Microservices
summary
Paper claims that run-time performance robustness certificates (PERCs) give statistical upper bounds on tail latency under bounded perturbations and that integrating them reduces SLO breaches. It builds a queueing-theoretic MM1 PSQ model, fits end-to-end latency to a gamma(alpha,beta), uses compressive-sensing gradient estimation to project worst-case alpha/beta, and implements Galileo to add rewards and shields to learned controllers. In 250+ hours on real benchmarks Galileo-backed autoscaling cuts violations to 0.05% SLO violations while costing about 20% higher peak CPU.
AI-written summary · about
chapters (15)
- 0:00Performance robustness intro
- 1:00Latency variability challenges
- 3:00Why controllers fail
- 3:31Performance robustness defined
- 4:35Galileo framework intro
- 5:17Queuing theory model
- 7:43Calibrating performance model
- 8:04Worst-case tail latencies
- 9:27Gradient estimation
- 10:51Validating controller actions
- 12:16Evaluating Galileo
- 13:00Perks for worst-case bounds
- 13:20Galileo reduces SLO violations
- 14:23Summary and implications
- 14:44Open-source and future
related talks
45mUnderstanding Microservice Performance
35mSCALA BASED SERVICES RESILIENCE IN THE REAL WORLD
25mBuilding a highly scalable system to assure Business Resilience
14mPerformance Optimization in High-Volume Transaction Systems
40mChaos Engineering at Scale
40mMeasuring Micro-services
30mMeasuring Microservices
49mScaling Reliability: So You Want to Add a 9