Tiancheng-Luo / csp-presentation

Parallelization of travelling salesman problem using PyCSP

Parallelization of travelling salesman problem (TSP) using PyCSP

Used as an example in the presentation: "Communicating Sequential Processes (CSP) - An alternative to the actor model"

The implementation models TSP as a complete, undirected weighted graph, and implements the branch-and-bound optimization. The parallelization uses a master-slave pattern, and requires Python 2.6 or 2.7.

How to run sequential version:

python run_sequential.py <num_cities>, where

• num_cities is number of cites/node in TSP

How to run parallel version:

python run_parallel.py <num_cities><task depth><num workers>, where

• num_cities is number of cities/nodes in TSP
• task depth is the length of sub-routes the master will use as tasks in the parallelization. A larger task depth results in more tasks.
• num workers is number of workers that will be spawned

Branches

• iter1: Does not use branch-and-bound optimization in parallelization. Less complexity.
• cluster: Minor modification for using the cluster module in PyCSP.