A tool to simplify benchmarking graph analytics platforms & graph databases

• rust
• python
• ansible

• terraform
• vagrant also requires plugins:
  • libvirt

1. Clone the repo

git clone https://github.com/graph-benchmarks/graph-benchmarks

Check the example.config.toml for an example configuration for the benchmark runner, with all options explained

2. Build everything

./build.sh

A graph benchmarking platform based on graphalytics

Usage: bench [OPTIONS] <COMMAND>

Commands:
  setup      Setup platform & kubernetes
  benchmark  Run benchmarks
  destroy    Teardown platform & kubernetes
  ls         List Resources
  dashboard  Port forward dashboard
  help       Print this message or the help of the given subcommand(s)

Options:
  -v, --verbose      Verbose logging
  -f, --file <FILE>  Configuration file [default: config.toml]
  -h, --help         Print help
  -V, --version      Print version

Use the environment variable LOG for debug logging, eg. LOG=info

3. Run setup

You might need sudo/root access

bin/bench setup

4. List resources

bin/bench ls

5. Run benchmarks

bin/bench benchmark

6. Destroy resources

bin/bench destroy

Datasets are placed in the datasets folder, with a vertex and edge file. eg:

├── twitter
│   ├── twitter.v
│   ├── twitter.e
│   ├── config.toml
├── facebook
│   ├── facebook.v
│   ├── facebook.e
│   ├── config.toml

The config.toml file contains these parameters:

weights = false
directed = false

Arguments to specific graph platforms can be provided in the config file and are driver specific.

[setup.graph_platform_args.graphscope]
cpu = 4
memory = "2Gi"

Specify cpu and memory for nodes using vagrant

[setup.master_platform_env]
cpu = "8"
memory = "8192"

[setup.worker_platform_env]
cpu = "8"
memory = "8192"

Arguments to [setup.worker_platform_env] prefixed with worker-{n}, eg. worker-0some_argument=3 to pass arguments to the object map used to spin up nodes

Arguments can be passed to platforms on a provider like shown below, for vagrant-libvirt

[setup.platform_args]
storage_pool_path = "/path/to/storage_pool"

Comes with drivers for:

• graphscope

Build arguments are specified in build.config.toml. If you want to build and run a minimal setup, remove unnecessary drivers & platforms from the build config file, and re-run build.sh

Follow one of the reference implementations for a more detailed guide.

1. Create a rust library with the name driver-config inside the driver folder
2. Write your driver code
3. Write a dockerfile to containerize your driver
4. Add a setup.yaml, and a remove.yaml for creating and destroying the graph platforms
5. The driver-config rust crate must implement the DriverConfig trait present in common/src/driver_config.rs
6. Add it to build.config.toml

The driver should take a single argument, to a yaml config file with the following format:

config:
  id: 32 # integer database ID given for each run, provided by the runner
  algo: sssp # the algorithm to run
  log_file: /path/to/log/file

postgres:
  host: postgres host
  port: 5432
  db: postgres db
  user: postgres user
  ps: postgres password

platform:
  host: host of graph platform
  port: port of graph platform

dataset:
  vertex: /path/to/vertex/file
  edges: /path/to/edge/file
  name: dataset name

1. Implement the Platform trait as a rust crate present in common/src/provider.rs