Distributed Simulation

Distributed Simulation#

Through the use of mango, it is possible to run ASSUME in a distributed way.

This is not supported on Windows for now.

There are different scenarios in which this might be wanted or needed.

  • Improve performance of large simulation by utilizing multiple cores, avoiding the GIL <https://wiki.python.org/moin/GlobalInterpreterLock>

  • Run simulation across multiple computer nodes to avoid exposing proprietary strategies

  • Run simulation with real time clock

  • Scalability simulation

The feature can be enabled for a single node using the -p command line parameter. This sets the distributed_role = True and opens a port for connections on 9010 As documented in Assume CLI.

World Distributed Role#

When creating a World module, one can specify the optional parameter distributed_role. This can have three values: True, False and None (default)

  • None: no distributed role behavior is established. The whole simulation is run in a single process, utilizing one core.

  • True: distributed behavior is used. Every Agent is created with its own mango container in a separate process. The mango containers communicate the current time between each other through the DistributedClockManager.

  • False: specifies the distributed_role as an agent which does not manage its own Clock but uses a DistributedClockAgent which connects to a manager_address and receives clock updates from it.

Distributed Example#

Besides the command line interface param, there is the option to run distributed simulations from the example.

The example contains the following structure:

  • main.py (contains script to run distributed simulation in single terminal using subprocesses)

  • world_manager.py (contains manager script)

  • world_agent.py (contains agent script)

  • config.py (contains common config used by the other modules)

This makes it possible to run the simulation either using:

python -m distributed_simulation.main

which starts the simulation in your terminal.

Or use two separate terminals, start the first one using:

python -m distributed_simulation.world_agent

The script tries to connect to the manager now. Next step is to start the manager. Open another terminal, activate your python environment and run

python -m distributed_simulation.world_manager

Now we are starting a simulation with two different terminals available.

Docker Example#

Of course it is also possible to run this using docker to emulate having different machines.

This can be done using:

docker compose –profile mqtt up

you should see a MQTT message broker being created as well as the simulation containers.

Contents