graph-bench¶

Comprehensive benchmark suite for comparing graph database performance.

GitHub

Overview¶

graph-bench provides standardized, reproducible performance testing across multiple graph database engines. It measures 25 benchmarks organized into 7 categories.

Supported Databases¶

Benchmark Categories¶

Storage (4 benchmarks)¶

• node_insertion: Bulk node creation
• edge_insertion: Bulk edge creation
• single_read: Individual node lookup
• batch_read: Batch node retrieval

Traversal (5 benchmarks)¶

• hop_1 / hop_2: N-hop neighbor traversal
• bfs / dfs: Breadth/depth-first search
• shortest_path: Dijkstra shortest path

Algorithms (4 benchmarks)¶

• pagerank: PageRank centrality
• community_detection: Louvain community detection
• betweenness_centrality: Betweenness centrality
• closeness_centrality: Closeness centrality

Query (3 benchmarks)¶

• aggregation_count: Count aggregations
• filter_equality: Equality filtering
• filter_range: Range filtering

Pattern Matching (2 benchmarks)¶

• triangle_count: Triangle enumeration
• common_neighbors: Common neighbor queries

Graph Structure (4 benchmarks)¶

• connected_components: Component detection
• degree_distribution: Degree statistics
• graph_density: Density calculation
• reachability: Path existence queries

Write Operations (3 benchmarks)¶

• property_update: Property modifications
• edge_add_existing: Edge creation between existing nodes
• mixed_workload: Combined read/write operations

Test Scales¶

Installation¶

uv add graph-bench

Usage¶

CLI¶

# Run all benchmarks for Grafeo at small scale
uv run graph-bench run -d grafeo -s small --verbose

# Run specific benchmark
uv run graph-bench run -d grafeo -b traversal.bfs -s medium

# Compare multiple databases
uv run graph-bench run -d grafeo,neo4j,duckdb -s small

Python API¶

from graph_bench import BenchmarkRunner
from graph_bench.adapters import GrafeoAdapter

adapter = GrafeoAdapter()
runner = BenchmarkRunner(adapter, scale="small")
results = runner.run_all()

for result in results:
    print(f"{result.name}: {result.mean_ms:.2f}ms")

Server Setup¶

For server-based databases, use Docker Compose:

cd graph-bench
docker compose up -d

Metrics Collected¶

Each benchmark captures:

• Timing: mean, median, min, max (nanoseconds)
• Throughput: operations per second
• Percentiles: 95^th, 99^th percentile latencies
• Stability: standard deviation, coefficient of variation

Requirements¶

• Python 3.12+
• Docker (for server databases)
• 8GB+ RAM recommended for large scale

License¶

Apache-2.0