SLATE: Service Layer Traffic Engineering
summary
SLATE is a service-layer traffic engineering system that optimizes multi-cluster microservice request routing and runs as a web-service controller plus an Envoy sidecar WASM proxy plugin on Kubernetes with Istio. It combines a model-based global optimizer (uses topology, per-class latency models, placement) with a local explorer called performance-aware policy adjustment (Papa) to fix model mismatch. Evaluated on GCP (Oregon, Utah, Iowa, South Carolina) with ~500 CPU cores, SLATE cuts average latency ~4× (up to 18.3×), tail latency up to 3.6×, and egress cost up to 2.64×, converging after runtime changes in ~30–60 s.
AI-written summary · about
chapters (9)
- 0:00SLATE and microservices routing
- 1:21Request routing vs load balancing
- 2:47Traffic engineering opportunities
- 6:56SLATE system design
- 7:57Global optimizer challenges
- 10:23Performance-aware policy adjustment
- 13:11Evaluation and benchmark results
- 15:21Latency and cost improvements
- 17:05Conclusion and open-source
related talks
12mSLATE: Globally Optimizing Request Routing
18mLessons from Scaling AI Powered Translation Services Using Istio
56mObservability in Action for your Service Mesh
38mThe Untapped Power of Services - L7 Load Balancing Without a Service Mesh
28mSharding a GraphQL Gateway for Blast Radius Reduction
27mService Mess to Service Mesh
35mIntroduction to service mesh with Istio and Kiali
09mK8s API Gateway: Service Mesh Strategies for Scalable Solutions