Distributed physics experiments

Experiments on distributed physics with rapier. Note that this repository is not production-ready and involves many manual steps in order to deploy the rapier distributed simulation cluster. The main goal of this repository is to experiment with designs for building a distributed physics systems aiming to simulate 1 million objects simultaneously.

Several executable are involved:

• steadyum-partitionner: responsible for partitioning the 3D simulation domain, assigning work to runners, and (virtual) time synchronization between runners.
• steadyum-runner: responsible for running a Rapier physics engine instance. It receives objects to simulate, simulates them, and automatically detects if an object needs to be transferred to a different runner as it moves.
• steadyum-updater: this is not a mandatory component, but it convenient for fast iterations. When deployed on a new node, it will communicate with the master partitionner instance to automatically download the latest versions of the partitionnar and runner executables, and deploys them locally.

Building

Build each executable with cargo, just like any other rust project. We recommend stripping the executables to reduce their sizes:

cargo build --release -p steadyum-partitionner --features dim3
strip target/release/steadyum-partitionner

cargo build --release -p steadyum-updater --features dim3
strip target/release/steadyum-updater

cargo build --release -p steadyum-runner --features dim3
strip target/release/steadyum-runner

Deploying

Once all the executables are built you will need at least two nodes: one for running the master partitionner, one for running the runners.

Master partitionner node

On the master partitionner node:

1. Add a configuration file .env with the following content:

REDIS_ADDR="redis://127.0.0.1"
PARTITIONNER_ADDR="http://localhost"
PARTITIONNER_PORT="3535"
RUNNER_EXE="./steadyum-runner"
ZENOH_ROUTER=""

1. Run steadyum-partitionner.

Runners nodes

There can be one or many runner nodes. They will all contribute to the same physic simulation. On each of the runner node:

1. Add a configuration file .env with the following content. Be sure to replace the ip by the IP of the node running the master partitionner. Replace ens4 by the name of the private network interface of the runner node. This is used to automatically retrieve its ip address on the private network.

PARTITIONNER_ADDR="http://10.0.2.153"
PARTITIONNER_PORT="3535"
RUNNER_EXE="./runner"
PRIV_NET_INT="ens4"
ZENOH_ROUTER="tcp/10.0.2.153:7447"

2. Upload the steadyum-updater and run it: ./steadyum-updater. Based on the env file, it will automatically communicate with the master partitionner and download the necessary executables locally (runner and partitionner).