Benchmarks

This section provides performance benchmarks for all Neumann crates, measured using Criterion.rs.

Running Benchmarks

# Run all benchmarks
cargo bench

# Run benchmarks for a specific crate
cargo bench --package tensor_store
cargo bench --package relational_engine
cargo bench --package graph_engine
cargo bench --package vector_engine
cargo bench --package neumann_parser
cargo bench --package query_router
cargo bench --package neumann_shell
cargo bench --package tensor_compress
cargo bench --package tensor_vault
cargo bench --package tensor_cache
cargo bench --package tensor_chain

Benchmark reports are generated in target/criterion/ with HTML visualizations.

Performance Summary

In-Memory Operations

Durable Storage (WAL)

All engines (RelationalEngine, GraphEngine, VectorEngine) support optional durability via open_durable() with full crash consistency.

Hardware Notes

Benchmarks run on:

• Apple M-series (ARM64) or Intel x86_64
• Results may vary based on CPU cache sizes, memory bandwidth, and core count

For consistent benchmarking:

# Disable CPU frequency scaling (Linux)
sudo cpupower frequency-set --governor performance

# Run with minimal background activity
cargo bench -- --noplot  # Skip HTML report generation for faster runs

Benchmark Categories

Storage Layer

• tensor_store - DashMap concurrent storage, Bloom filters, snapshots
• tensor_compress - Tensor Train, delta encoding, RLE

Engines

• relational_engine - SQL operations, indexes, JOINs, aggregates
• graph_engine - Node/edge operations, traversals, path finding
• vector_engine - Embeddings, SIMD similarity, HNSW index

Extended Modules

• tensor_vault - Encrypted storage, access control
• tensor_cache - LLM response caching, semantic search
• tensor_blob - Blob storage operations

Distributed Systems

• tensor_chain - Consensus, 2PC, gossip, sparse vectors

Query Layer

• neumann_parser - Tokenization, parsing, expressions
• query_router - Cross-engine query routing