500M-row benchmark: hand-written BFS beats NebulaGraph in MySQL, 2.7ms vs 23.6ms
JeremyCMorgan · x · 2026-11-03
Developer Mehdi Mohseni published a reproducible graph-query benchmark loading the same graph into MySQL and NebulaGraph, scaling from 1M to 500M rows, with all code and raw results open-sourced on GitHub.
- Background: his production system once modeled a graph in a relational DB; at 500M rows it collapsed on a $15-20k/month AWS VM, with single-relationship queries taking 15 minutes. Migrating to ArangoDB, then Neo4j, then NebulaGraph brought queries down to a few seconds (7-10s at worst).
- Results: at 500M rows, a hand-written BFS shortest-path query in MySQL returned in 2.7ms, NebulaGraph took 23.6ms, and MySQL's recursive CTE timed out after 15 minutes.
- Key takeaway: the algorithm mattered more than the database — application-level graph traversal on a relational DB can outperform a dedicated graph database, though the benchmark also partly confirms the collapse story of pure relational modeling at scale.
More from Infra
- DeepSeek V5 reportedly to be pretrained on Huawei Ascend chips, Ascend 960DT due Q1 2027 — teortaxesTex · 2026-11-03
- Korean HBM smuggled into China: The Information investigates — AccBalanced · 2026-11-03
- Researchers dissect NVIDIA GPU confidential computing and test its security guarantees — burkov · 2026-11-03
- Nebius CTO: threw away the whole stack on day one with 1,600 H100s — now $17.4B Microsoft deal — AccBalanced · 2026-11-03
- Inference CEOs say they're doing SEO for AI agents that buy compute, not just humans — abhijithneil · 2026-11-03
- Fitting 4x RTX 3090 in one case: which PCIe 4.0 risers actually work — infectiousstupidity · 2026-11-03