Graviton2: How Honeycomb Reduced Infra Spend by 40%
summary
Honeycomb reports a 40% reduction in AWS spend by moving high‑volume services to Graviton2 arm64, measuring outcomes not promises. Benchmarks show ~10% lower median latency, much less tail latency, and CPU packing rising from ~60% to ~85%, while retriever gains up to triple effective throughput with more cores. Migration uses Go cross‑compile, CircleCI, Terraform Cloud, and a dog food → kibble pipeline; Kafka on Graviton2 causes a rollback and a documented incident.
AI-written summary · about
chapters (11)
- 0:00Graviton2 migration intro
- 1:43Reliability goals with SLOs
- 4:11Multi-environment testing strategy
- 5:34ARM64 infrastructure prep
- 6:36Dogfood A/B testing Graviton2
- 7:37Shepherd production migration
- 7:58Retriever performance gains
- 9:01Kafka migration challenges
- 10:03Key takeaways
- 11:48Team well-being during migrations
- 12:29Final thoughts
related talks
27mOptimizing price performance with feature management and observability (PRT259)
32mUsing Serverless Functions for Real-time Observability
23mBuilding & Running a Scuba-Inspired Observability Tool
40mManaging to Your SLO Amidst Chaos
51mThe Quest for Speed: Journey to 50% Better P99 Times with Go
26mGraphite@Scale or How to Store Millions of Metrics per Second
48mCloud Chaos & Microservices Mayhem
26mAndromeda: Performance, Isolation, and Velocity at Scale in Cloud Network Virtualization