NEST-simulator Beowulf cluster

The deloy system requires at least python 2.7 and DigitalOcean python-bindings (pip install -U python-digitalocean). It was tested up to 1/2 million neurons (Brunnel balanced network) on 400vcpus with 1.28To RAM for ~19$ hourly. This script is provided as is and comes with no warranty nor liability.

Motivation

This was part of a bachelor semester project carried out at the Laboratory of Computational Neuroscience from Swiss Institute of Technology in Lausanne. The command was used to evaluate different approaches of implementing spike-timing dependent plasticity on various size and networks. You can find the resulting work here.

Subcommands

Digital Ocean token is stored in .key file at command root. Be sure to never share it with anyone.

• ./clusterify.py key [token]: setup Digital Ocean token (empty for removing it)
• ./clusterify.py list: list current cluster
• ./clusterify.py create [2|3|...] [1gb|2gb|...]: create n-cluster of given size and setup ssh everywhere
• ./clusterify.py install [script]: execute given bash script on each node
• ./clusterify.py run [1..] [program]: run given python (PyNEST) program on the cluster using mpi and n processes
• ./clusterify.py delete: delete: remove current cluster

Common scenario

Global installation takes 15 minutes on average.

./clusterify.py key digitalOceanToken
./clusterify.py create 4 8gb
./clusterify.py list
./clusterify.py install install_env.bash
./clusterify.py install install_nest.bash
./clusterify.py install install_stdpmodule.bash
./clusterify.py run 16 brunnel.py
./clusterify.py run 8 brunnel.py
./clusterify.py run 4 brunnel.py
./clusterify.py delete