EmberGraph (pre-published)

To query a pattern on multi directed graph with label / attribute filtering efficiently, support Neo4j and SQLite3 as the storage layer.

❤️ Please give a Star / Follow if you like this project

To begin with

The project is still under development, however it has already been proved to be a correct implementation.

The test dataset is LDBC-SNB-SF0.1, and the example used for testing all come from LDBC-SNB Business Intelligence (BI 1 ~ 20). You could find the actual query statements from ./resources/cypher.

EmberGraph is only one-more-step to be formally published:

To optimize the matching order generation algorithm (most likely to be iterative dynamic programming just like what Neo4j has done since it's 2.2 version)

What's the most impressive?

Even though we're still working on the one-last-step to use the better matching-order seeking strategy, the performance of this pre-published version is approximately equivalent to that of Neo4j. (Oh, even faster in some cases!)

How to build

You should have installed git-lfs first, to correctly clone this project:

git clone https://github.com/DrEden33773/EmberGraph
git lfs pull
cargo build # This will automatically install uv if in need, and then initialize `./ember-graph-import`

Then, you should manually do some initialization steps:

1. Transformed the original dataset into a Neo4j-import-friendly one. (Using ./ember-graph-import/raw_data_formatter.py)
2. Generated basic label-based statistics. (Using ./ember-graph-import/statistics_generator.py)
3. Imported Neo4j-import-friendly dataset into Neo4j or SQLite3. (Using ./ember-graph-import/neo4j_admin_import.py or ./ember-graph-import/sqlite_import.py)

All of the python scripts mentioned above could executed via the command:

cd ./ember-graph-import
source ./.venv/bin/activate
uv run <SCRIPT-NAME>

Right now, if you want to, you could run commands:

cargo run --example bi_<x> # (where x in [1..=20])

To check the query result of bi_1 to bi_20.

Something important for release mode building

Yes, you might have guessed -- It's totally possible to get the highest performance to build under the release mode.

However, I have to mind you that could be TOO SLOW.

So, if you really don't mind, here're several better options:

• Linux:

cargo build --release --all-target -j $(nproc)

• Mac:

cargo build --release --all-target -j $(sysctl -n hw.ncpu)

• Windows(Powershell):

cargo build --release --all-target -j $env:NUMBER_OF_PROCESSORS

• Windows(CMD):

cargo build --release --all-target -j %NUMBER_OF_PROCESSORS%

After all, that's a short-term pain for long-term gain😂.

Then, you could run the command below to check the query result of bi_1 to bi_20:

cargo run --release --example bi_<x> # (where x in [1..=20])