graphsub.com / wiki / graph-database-benchmarks
Graph Database Benchmarks
Graph database benchmarks compare lookup latency, traversal latency, throughput, memory footprint, analytics support, and operational workflow fit.
What to measure
Useful graph database benchmarks include node lookup P99, edge traversal P99, saturated throughput, memory per million nodes, vector search support, and multi-hop query performance.
GraphSub performance lane
The GraphSub product map highlights 67us node lookup, 125us traversal, 25M ops/s, and 2.1GB per 1M nodes for the GraphSub lane.
Beyond speed
For agent systems, benchmark value also depends on memory, replay, source links, approval traces, and whether the graph can support operational workflows.
Code example
Capsule planner routesbash
curl 'https://graphsub.com/api/plan?capsule=trading-control-graph&target=enterprise-team&workload=risk-recompute' \
| jq '.execution_preview | {capsule, buyer, workflow, artifacts}'
curl -X POST 'https://graphsub.com/api/execute?capsule=regulated-evidence-graph&workload=audit-evidence' \
| jq '.execution.replay'
curl 'https://graphsub.com/api/examples'Use the live Modal routes to build and execute a capsule plan.
FAQ
- What is a fair graph benchmark?
- A fair benchmark publishes hardware, versions, dataset shape, query mix, concurrency, cache state, and P50/P95/P99 latencies.
- Why include workflow fit?
- Agent systems need more than fast graph queries; they need explainable state and replayable outputs.