Design of a Composable Infrastructure Platform
summary
DriveScale Composer composes secure bare-metal application clusters from disaggregated compute and storage at production scale (thousands of servers and ~10,000 drives, design target up to 100,000 managed elements). Platform uses triply replicated management with ZooKeeper+MongoDB to store atomic JSON cluster configurations, lightweight agents on every node that retry forever to reach the desired state, and standard Linux plumbing (iSCSI, NVMe over Fabrics, multipathing). Benchmarks show throughput near direct‑attached SSDs with modest latency tail effects; rolling upgrades and non‑disruptive rebinds are supported.
AI-written summary · about
chapters (11)
- 0:00Introduction to composable
- 1:04Defining composable infrastructure
- 4:07Application use cases
- 7:26Managing infrastructure challenges
- 11:06Scaling and maintenance
- 13:04User experience and demo
- 16:12Key concepts and inventory
- 19:22System architecture
- 21:55Distributed configuration
- 25:00Atomic writes and recovery
- 27:22Performance benchmarks
related talks
21mBuilding Distributed Platforms at Extreme Scale on Fast Changing Data
47mWeb-Scale IT Engineering.
43mSystems @Scale Summer Q&A Week 2
26mGuide through the galaxy - Structure of modern information systems
32mData @Scale Seattle- Naresh Sundaram
31mAn Adventure in Multi Cloud with Kubernetes and Distributed Data
01h 08mDystopia as a Service
02mStarcluster