Simple Benchmark for the Rust Extensions of the Neo4j Python Driver

This repository contains a simple benchmark for the Rust extensions of the Neo4j Python driver. The benchmark compares the performance of running different workloads through the Python driver with and with having the Rust extensions installed.

Workloads

The workloads are chose to test:

• Read-heavy work (called deserialization)
• Write-heavy work (called serialization)

Further, each workload is parameterized along two dimensions:

• Width (how large a single unit of data is)
• Length (how many units of data there are)

In particular, this means:

• Deserialization of NxM: The query produces M records with 1 column containing a list of N pieces of data.
  In this case, the data was chosen to be the string "qwertyuiopasdfghjklzxcvbnmQWERTYUIOPASDFGHJKLZXCVBNMM".
• Serialization of NxM: The query is submitted with a parameter that's an N long list of lists with M elements each.
  In this case, the elements were chosen to be the string "qwertyuiopasdfghjklzxcvbnmQWERTYUIOPASDFGHJKLZXCVBNMM".

Each workload is repeated multiple times simply summing up the time take to get more stable results.

Results

Here are the results of running the benchmark on my machine (Lenovo p1 Gen. 3, Intel i9–10885H, 64 GB DDR4 RAM@3200 MHz):

For a discussion of the results, see the blog post at https://medium.com/neo4j/neo4j-python-driver-10x-faster-with-rust-7059c56e8674.

Run It Yourself

• Get docker
• Get Python (3.12) (including pip and venv/virtualenv)
  run.sh assumes that the Python executable is called python3.12, if it's not, you may simply adjust the script
• Start Neo4j in docker:

  docker run --rm -p 7687:7687 --name=neo4j-rust-ext-bench \
    -e NEO4J_AUTH=neo4j/pass \
    -e NEO4J_dbms_security_auth__minimum__password__length=1 \
    neo4j:5.23

• Get bash (or do the steps in run.sh manually, skipping the next step)
• Run ./run.sh
• Find the results in ...

Developing

Before committing, please run

pip install -Ur requirements.txt
ruff format && ruff check --fix && mypy --strict --ignore-missing-imports .