Overcoming Distributed Databases Scaling Challenges with Tablets
summary
ScyllaDB replaces vnode/ring fragility with dynamic 5 GB tablet shards that split, move, and compact independently under Raft control, letting many replicas stream to a joining node in parallel. A cluster hosts tens of thousands of tablets, balances automatically via a tablet scheduler, and supports mixed node sizes. Benchmarks show adding three nodes in parallel while doubling load from 50k to 100k req/s keeps 1–2 ms P99 latency and sustains ~1 GB/s compaction per node on 8‑vCPU machines.
AI-written summary · about
chapters (7)
- 0:00Introduction and background
- 1:05Engineering for performance
- 3:26The evolution of sharding
- 11:06Introduction to tablets
- 17:39Tablet scheduling and scaling
- 23:47Demonstration and test results
- 28:34Conclusion and wrap-up
related talks
11mScyllaDB Tablets in Action
41mPerformance & Fault Tolerance: Building a Modern Database
29mEnterprise Scale: How a Distributed Architecture Can Deliver Scale and Resilience
59mScaling Raft
29mData Modeling for Distributed SQL
59mServerless nature of
35mHow to Save on Scaling by Moving from Cassandra to Scylla DB
36mHorizontal Scaling with Vitess