OLTP design debates in the wild: MVCC, locks, and distributed strategies

MVCC vs locks isn’t academic—it’s how you keep balances safe under real-time pressure. TigerBeetle leans into write-heavy workloads with strict serializability, trading SQL-style flexibility for raw throughput when contention spikes ^[1].

OLTP Concurrency Showdown In TigerBeetle, 1000-2000x more performance than a single-node OLTP DBMS is possible under contention, while still enforcing strict serializability ^[1]. That design contrastes with common MVCC implementations where locking is central. In PostgreSQL, MVCC is widespread, but most commercial MVCC systems still rely on row locks; OCC is comparatively rare ^[1]. This highlights a clear trade: locking can boost robustness across varied workloads, but MVCC strategies may surface different bottlenecks in high-contention, low-latency apps ^[1].

Distributed SQLite in the wild • SQLite clusters that are leaderless can run on an embedded NATS JetStream server, exposing an HTTP API or the PostgreSQL Wire Protocol ^[2]. • The ha-sync SQLite extension enables live local read replicas, easing distribution without sacrificing local latency ^[2]. • CDC is part of the approach, helping maintain consistency across nodes in a distributed setup ^[2].

Closing thought: there isn’t a one-size-fits-all; DB engines choose concurrency controls and distribution layers that trade throughput, correctness, and ops complexity in practice ^[1][2].

References