Cover image for LioranDB V2 Pre-Alpha Benchmark Summary

Swaraj Puppalwar

Over the past few weeks I've been benchmarking LioranDB V2, a document database engine written completely from scratch in Rust.

Instead of publishing screenshots or selected numbers, I've decided to publish the raw benchmark logs so anyone can inspect the results themselves.

This is Pre-Alpha software. The goal isn't to claim perfection, but to be transparent about where the database stands today and where it's heading.

Test Configuration

  • Language: Rust
  • Architecture: Distributed Document Database
  • Nodes: 4
  • Partitions: 8
  • Threads: 32
  • Batch Size: 256
  • Dataset Sizes:
    • 1 Million Documents
    • 10 Million Documents

Features enabled during benchmarks:

  • Secondary Indexes
  • Full Text Search
  • WAL (Write Ahead Log)
  • MVCC
  • Parallel Reads & Writes

Write Performance

The write benchmarks were executed using realistic document workloads while maintaining durability through the WAL.

Results reached approximately

  • 23K to 25K writes/sec during the benchmark
  • Sustained write throughput across millions of documents
  • Stable WAL group commit behavior

Benchmark logs:

Read Performance

Read benchmarks include

  • Primary Key Lookups
  • Secondary Index Queries
  • Range Queries
  • Pagination
  • Full Text Search

During the benchmarks the engine demonstrated

  • Low millisecond latency
  • High parallel throughput
  • Efficient index lookups
  • Stable read performance under concurrency

Benchmark logs

Mixed Workload

Real applications rarely perform only reads or only writes.

The soak benchmark continuously executed mixed workloads consisting of writes together with concurrent reads.

Target workload:

  • ~10K Writes/sec
  • ~25K Reads/sec
  • ~35K Total Operations/sec

The goal was to evaluate long-running stability rather than peak numbers.

Benchmark log

https://www.liorandb.com/benchmark_ldb/soak/soak_10m.txt

Crash Recovery Test

Reliability matters more than benchmark numbers.

LioranDB V2 was repeatedly terminated during active writes and restarted to verify WAL recovery.

The crash benchmark performs multiple forced crashes while documents are still being inserted.

Recovery validates:

  • WAL replay
  • Metadata consistency
  • Duplicate prevention
  • Missing document detection

Benchmark log

https://www.liorandb.com/benchmark_ldb/crash/crash_test_1m_9c.txt

What These Benchmarks Show

Current pre-alpha results demonstrate:

  • Multi-node architecture
  • Parallel execution
  • Durable WAL commits
  • Secondary indexes
  • Full-text search
  • Sustained write throughput
  • Concurrent reads and writes
  • Crash recovery support

The project is still under active development, and these benchmarks are intended to provide transparent insight into its current capabilities rather than serve as marketing claims.

Benchmark Repository

All raw benchmark logs are publicly available.

https://www.liorandb.com/benchmark_ldb/

About LioranDB

LioranDB is an open-source distributed document database written from scratch in Rust.

The project focuses on building modern developer infrastructure with performance, simplicity, and transparency as core principles.

Website

https://www.liorandb.com

About Lioran Group

LioranDB is developed by Lioran Developer Solutions (LDS).

Website

https://www.lioransolutions.com

Company

https://www.lioran.group

About the Author

Swaraj Puppalwar

Founder & CTO, Lioran Group

GitHub

https://github.com/UltronTheAI

Feedback, benchmark reviews, and contributions are always welcome. Every benchmark log is public so anyone can inspect the results, reproduce the tests, and help improve the project.