Running A Distributed Simulation (Example)

In order to run the simulation environment:

1. Create 10 Experiment Files (This defines the names and number of simulation experiments to run)

export EXPERIMENT_DIR=/home/$USER/experiments
cd $EXPERIMENT_DIR
touch experiment{01..10}.exp

Note

Experiment names should not contain underscores (‘_’), and must have the extension ‘.exp’.

2. Set a Reservation Directory (JobAllocator communicates with the matlab processes by writing files to this directory [1]):

export RESERVATION_DIR=/home/$USER/reservations/

Note

It is important that RESERVATION_DIR is accessible from all runnning instances of the distributed code. This is why it is advisable to run it on the home directory, which at D-ITET is accessible from all machines.

3. Launch JobAllocator

python job_allocator.py --dir $RESERVATION_DIR --databook $RESERVATION_DIR/obj/databook.pkl --t1

Note

For more information on JobAllocator switches and options, run python job_allocator.py --help

4. Deploy and run jobs on the cluster

./job_submission.sh

[1]

JobAllocator and the matlab process currently communicate with each other by writing files to this directory. Note that this mechanism is a temporary workaround. Initially, the communication was meant to happen over HTTP through REST calls to JobAllocator (which services API calls), as documented in the report.