Data Graphs benchmark vs Neo4j
Data Graphs

Data Graphs Cypher Benchmark - 110x faster than Neo4j

July 24, 2024
3 mins
We benchmarked Data Graphs and it’s fast. Really fast. In fact 110x faster than Neo4j on average, but that’s not the whole story. It gets better.

Reasoning

We recently completely overhauled our Data Graphs Cypher Query engine from the ground up. We threw away the book on graph-querying. A new novel algorithm for ultra-efficient processing of GQL and Cypher graph queries.. The results were good. Really good. Obviously we wanted to show these off, so what better way to do this than using one of the standard graph benchmarks; benchgraph.

Headlines

  • Across 101 x 5 query invocations, Data Graphs completed the entire benchmark in 1.6 minutes compared to 3 hours for Neo4j - on average 110x faster.
  • Data Graphs was faster on 94 out of the 101 queries. Those queries where Neo4j was faster were the simplest very fast aggregation queries where the difference in query times were just a few milliseconds.
  • On one of the deepest and broadest graph queries, Data Graphs was 518x faster, completing in just under 1 second compared to Neo4j at 517 seconds ! 
  • The more “graphy” the query is, the faster Data Graphs is compared to Neo.

The Benchmark 

The benchmark is a set of 101 Cypher queries of varying complexity that are invoked 5 times each with average millisecond times calculated for each query. The graph used for the benchmark was the pokec medium dataset comprising 100k nodes and 1.77m edges, which is a genuine representation of an Slovakian social network.
https://github.com/memgraph/memgraph/tree/master/tests/mgbench#pokec

It’s clearly important that the benchmark is compared on identical hardware. We ran our tests on a 2.6GHz 6-Core Intel i7 with
16GB 2400MHz DDR4. 

We compared Data Graphs 2.0.25 against Neo4j 5.19.0
The results in full detail are available below.

Conclusions

This is a big deal. One of the key takeaways is that the more “graphy” a query is, in other words the deeper, broader, and more connected the graph query projection is, the better Data Graphs performs, and not by small margins but by orders of magnitude. Given this is the core purpose of a graph database it is surprising that Neo4j struggles in comparison, but is also a reflection of our advanced new query algorithm. Approximately for every graph-hop deeper the query pattern is, Data Graphs will out-perform Neo by one order of magnitude. The slowest query in the benchmark is a 4 hop path-pattern. We outperform Neo by 500x. We are confident on a 5-hop path we would out-perform by possibly 5000x.
This is genuinely ground breaking ! Watch this space as we have more to say on this.

When our Cypher query API is combined with our best-in-class query builder user interface, that makes it easy for non-experts to build queries, it really is a perfect combination. Videos and and a blog on the query UX will follow.
Get in touch now to try it out for free!

Detailed Benchmark Results