Comparing MongoRocks, WiredTiger and mmapv1 for performance and efficiency
summary
Comparative benchmarks on MongoDB 3.4 show MongoRocks (RocksDB engine) dramatically reduces space and write work: databases shrink to ~37GB vs 49GB for WiredTiger, and WiredTiger writes 3–4x more bytes per insert while mmapv1 writes ~2x more. Non‑unique secondary index updates are read‑free on MongoRocks, cutting read I/O and letting it sustain 16k inserts/sec on a 48‑core NVMe server where WiredTiger falls to ~5k and loses half its throughput on slower TLC SSDs.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:50MongoRocks background
- 3:58Space and write efficiency
- 6:04MongoRocks vs B-trees
- 9:24Read efficiency and workload
- 11:12Performance evaluation
- 13:21Insert Benchmark
- 14:27Small server results
- 16:53Concurrent inserts and queries
- 18:21Large server comparison
- 20:12Response time variance
- 21:37Slower storage impact
- 22:42Single vs multiple tables
- 23:45MongoRocks in production
- 24:51Compression sensitivity
- 26:58Q&A
related talks
57mRocksDB
01h 03mWiredTiger
46mUnderstanding and tuning WiredTiger
52mRocksDB Performance and Reliability Practics
46mUnderstanding and tuning WiredTiger
51m5 boring performance improvements in MongoDB 3.4 that matter
37mMoving on from RocksDB to something FASTER
24mPerformance tuning in BlueStore & RocksDB