Benchmarking SurrealDB 3.x vs. Postgres, Mongo, Neo4j and Redis (With Fsync)

Benchmark results released in March 2024 show that SurrealDB 3.x delivers significant performance improvements over previous versions and competes closely with, or surpasses, established databases like Postgres and MongoDB when tested under production-grade conditions with full disk durability enabled.

The benchmarks were conducted on identical hardware: an AMD Ryzen Threadripper 9970X, 128 GiB DDR5 RAM, NVMe storage, running Ubuntu 24.04. Every database was tuned for production, with settings like connection limits, buffer pools, and durability configurations optimized to reflect real-world deployment.

The workload involved 128 clients issuing 48 concurrent queries each on datasets of 5 to 15 million rows, including mixed data types and complex objects. All tests used full fsync, ensuring data durability on disk after each transaction, which slowed performance but provided an honest comparison.

Results reveal that SurrealDB 3.x’s CRUD throughput increased by approximately 31% over 2.x, reaching 141,000 operations per second, and full-table scan speeds improved by over 11,800%. Latency reductions ranged from 27% to nearly 100% depending on the query type. Notably, the new query planner and storage engine eliminated per-row decoding overhead, drastically reducing scan times.

Compared to other databases, SurrealDB outperformed Postgres and MySQL in write operations, with 1.5x faster creates and updates, and 1.8x faster deletes. It also surpassed MongoDB in read performance by about 1.3x and was significantly faster in unindexed filter scans, roughly 2.7x faster than MongoDB. Neo4j and Redis benchmarks are also included, but detailed results are still emerging.

Why It Matters

This benchmarking effort is important because it provides an apples-to-apples comparison of SurrealDB 3.x with well-established databases under realistic, production-like conditions. The results suggest that SurrealDB has made substantial progress in both transactional throughput and query efficiency, challenging the dominance of traditional relational and document stores in certain workloads.

For developers and organizations, these findings may influence database selection, especially for applications requiring multi-model capabilities, full durability, and high concurrency. The improvements could reduce infrastructure costs and simplify architecture by consolidating workloads into a single database engine.

However, it remains to be seen how SurrealDB performs in other scenarios, such as distributed setups, replication, and complex graph queries, which are not covered by these benchmarks.

Background

Previous benchmark rounds with earlier SurrealDB versions showed modest performance gains, but the latest results highlight a major internal overhaul of query and storage layers in SurrealDB 3.x. The benchmarking methodology was rigorous, using identical hardware, open-source testing harnesses, and production-grade configurations, including full fsync enabled for durability.

Historically, databases like Postgres and MongoDB have been the benchmarks for relational and document models, respectively. SurrealDB aims to unify multiple data models and provide transactional guarantees, which this benchmarking confirms can be achieved without sacrificing performance. The comparison with earlier SurrealDB versions demonstrates that the recent internal improvements have been substantial.

While the benchmarks focus on single-table workloads with large datasets, they do not yet address distributed deployments, replication, or complex graph queries, which are also key use cases for SurrealDB. The results are a snapshot of performance on a single, high-end server setup.

“The improvements in query planning and storage engine have reduced scan times by over 11,800%, enabling workloads that previously took minutes to complete in seconds.”

— SurrealDB team

“Running all databases with full fsync enabled provides a realistic picture of performance in production environments where data durability is critical.”

— Hacker News source

What Remains Unclear

It is not yet clear how SurrealDB 3.x will perform in distributed or multi-node configurations, or under different hardware setups. Further testing is needed to evaluate scalability and replication performance.

What’s Next

Next steps include testing SurrealDB in distributed environments, assessing replication and clustering capabilities, and benchmarking complex query workloads such as graph traversals. The SurrealDB team is expected to release further updates and detailed documentation on these aspects.

Key Questions

How does SurrealDB 3.x compare to traditional relational databases in real-world applications?

While benchmark results show strong performance, real-world performance depends on workload specifics, deployment architecture, and data complexity. SurrealDB’s multi-model capabilities may offer advantages in certain use cases, but comprehensive testing is needed for definitive comparisons.

Will SurrealDB support distributed or multi-node setups?

The current benchmarks focus on single-server performance. The SurrealDB team has indicated plans to develop distributed and clustering features, but these are not yet available or tested in the current release.

What are the main improvements in SurrealDB 3.x compared to earlier versions?

Major internal changes include a redesigned query planner and storage engine that eliminate per-row decoding overhead, leading to faster scans and lower latency. CRUD throughput has also increased significantly across all workloads.

Source: Hacker News