Graph traversal on object storage

How fast HydraDB answers multi-hop queries, what each extra hop costs, and how far it scales across cores. Measured against MinIO on a single 15-core machine.

hover to scrub

What each hop costs

Time for one query at each traversal depth, by supernode fanout and worker count. Depth h covers every path of length 1 to h.

Show the numbers

How far it scales

Throughput as workers are added, against perfect linear scaling.

Show the numbers

What a write costs

Time to commit one Cypher write, and what happens to that time when writers are added. Each write goes to its own id range, so it lands in the graph without touching the supernode.

Show the numbers

What the cache is worth

First query after a wiped cache, against steady state.

Show the numbers
Harness
bench/ in this repo, calling the engine in-process.
Storage
MinIO in Docker on the same host.
Graph
A supernode whose fanout edges each start a chain, one vertex per hop. Every query starts at the supernode. Fanout 50 to 10,000.