Latency, Throughput & Fault Tolerance: Building a Modern Database
summary
Shows a concrete migration and measurable wins: migrated 177 Cassandra nodes to 72 CLAB nodes, P99 40–125ms → 15ms, and claims 1.1M ops/s at <1ms P99 on AWS I4i. Describes the actual stack choices: a C++ low‑level, hardware‑adjacent engine using a shard‑per‑core / Seastar‑style framework to avoid JVM GC pauses and provide predictable latency, plus replication, tunable per‑query consistency, and multi‑DC topology for availability and locality.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:47Discord’s database migration
- 7:52Modern database history
- 11:51CIAB development and features
- 16:49CIAB architecture
- 20:23Data replication and availability
- 23:15Consistency levels and use cases
- 28:36Multi-data center databases
- 31:50NoSQL and CIAB resources
- 35:06CIAB Cloud service
- 36:29Q&A
related talks
41mPerformance & Fault Tolerance: Building a Modern Database
48m8ms/99th write percentile latency - is it fast?
45m8ms/99th write percentile latency - is it fast? (...)
01h 05mRoundtable: Cloud Databases as a Service
39mMaking Cassandra Faster in Cloud-native architecture
35mReplacing Cassandra: A Digital Transformation for the World’s Largest Digital Identity Network
45mCassandra Architecture: Understanding, Scaling and Optimizing
57mDSS Keynote — Back to the Future: Prepare Your Data Infrastructure for Anything