Non Functional Requirements For The Win!
summary
Clear difference between SLI (what is measured), SLO (P95 + max as the practical goal) and SLA (external contract with consequences); load matters—design load must accompany any latency target, e.g. 200 ms SLO vs 300 ms SLA. Tools used for induction and measurement: K6 or JMeter and dashboards. Kubernetes specifics: right-size pods by vertical testing, expect Java Spring Boot to need ~200–500 millicores light, 1–2 cores at scale, watch GC pauses, and set requests = limits to avoid node OOM kills.
AI-written summary · about
chapters (8)
- 0:00Introduction and speaker background
- 10:02Understanding nonfunctional requirements
- 20:56SLI, SLO and SLA explained
- 31:27Design load and its importance
- 43:08Reads vs writes in system design
- 50:35Availability and parallel systems
- 55:28Right-sizing pods and resources
- 1:00:01Summary and Q&A
related talks
37mReal World SLOs and SLIs: A Deep Dive
23mSLO's for Openstack - the API for Engineering Teams
37mUsing SLOs for Continuous Performance Optimizations of Your K8s Workloads
25mAn SLO-Driven Approach to Enhance Kubernetes Cluster Reliability
31mServerless Observability: a case study of SLOs
32m9 Things You Should Do When Starting to Use SLOs
28mLean Product Development Through SLOs
35mAvoiding Goodhart's Law - Use SLO's as Tools Not Cudgels