Orbit

Introduction

Given a space X, a list of generators f_1,...,f_n : X -> X and an initial vertex x_0 : X, the /orbit problem/ is to compute the least subset Orb of X such that Orb contains x_0 and is closed under all generators.

The orbit implementation presented here is specialised to the case where the space X is a finite subset of the natural numbers. For the purpose of benchmarking, the generators perform some deliberately irregular computation (by calling a naive recursive implementation of Fibonacci).

ARCHITECTURE

This orbit implementation operates on a distributed hash table. It follows a master/worker architecture, where each worker hosts a chunk of the hash table; the master simply initiates the computation and waits for termination.

How to build and run Orbit

There are two ways to run the benchmark. .

• Run on the local node (nonode@nohost):

    $ git clone git://github.com/amirghaffari/Orbit
    $ cd Orbit
    $ ./run
  

The config file for a local run is bench.config. After finishing the benchmark, the results are shown on screen.

• To run the benchmark on a cluster, you need to specify the cluster information (such as number of nodes, nodes name, number of experiments,and path where Erlang has been installed) in both files: run.sh and experiment.sh

    $ git clone git://github.com/amirghaffari/Orbit
    $ cd Orbit
    $ ./run.sh
  

The config file for clusters is template.config. After finishing the benchmark, the results are saved in results directory.