MongoDB Performance Improvements

4/2/2025

Introduced a "performance army" approach to systematically improve MongoDB performance, building on a tiger team's initial work. Key technical contributions include the optimization of simple query paths by bypassing the general query planner for IDHACK, leading to the development of the ExpressPlan code path. Replication latency was reduced by changing the acknowledgment strategy to occur after journaling writes, before in-memory application. A new regression detection mechanism using instruction counting with Linux perf_event_open syscall was implemented to catch minor performance degradations. The broader "army" phase saw optimizations in query yielding, clustered collections for config.transactions, locking mechanisms, authorization checks, memory allocation (TCMalloc), and improvements in common code paths like namespace string handling and generated parsers.

10/2/2024

This post announces MongoDB 8.0 and highlights its performance improvements, including a 36% read and 32% write speedup over 7.0. It details architectural optimizations leading to reduced memory usage and query times, and more efficient batch processing. Specific metrics include 36% better read throughput, 56% faster bulk writes, 20% faster concurrent writes during replication, and 200% faster complex time series aggregations. It also mentions sharding enhancements for faster data distribution and lower cost, and expanded Queryable Encryption for range queries. The post also notes the internal "dogfooding" experience with a 75% drop in query latencies for their build system.