GridEngine

A lightweight Python library for distributed computing on Sun Grid Engines

Introduction

GridEngine streamlines the process of managing distributed computing on a Sun Grid Engine. It was designed for iterating over algorithm and experiment design on a computing cluster.

GridEngine was intentionally designed to match the API of the built-in multiprocessing.Process and threading.Thread classes. If you have ever used these, the gridengine.Job class will be familiar to you.

At its core, gridengine is designed to transparently schedule and execute Jobs on a Sun Grid Engine computing cluster and return the results once the jobs have completed. All scheduling and communication whilst jobs are running are handled by gridengine.

The component layout of gridengine can be visualized as follows:

        |       JobDispatcher  ------>  Scheduler
Host    |           /\                    /
        |          /  \                  /
                  /    \                /
Comms   | ZeroMQ /      \              / Sun Grid Engine
                /        \            /
Cluster |     Job0  ...  Job1  ...  JobN

Jobs are wrappers around a function and its arguments. Jobs are constructed on the host and executed on the cluster. The JobDispatcher is tasked with collating and dispatching the jobs, then communicating with them once running. The JobDispatcher passes the jobs to the Scheduler to be invoked.

There are two schedulers:

• ProcessScheduler which schedules jobs across processes on a multi-core computer (laptop, etc). This is handy for debugging and experiment design before scheduling thousands of jobs on the cluster
• GridEngineScheduler which schedules jobs across nodes on a Sun Grid Engine (cluster). This scheduler can be used in any environment which uses DRMAA, it is not strictly limited to SGE.

Once the jobs have have scheduled, they contact the JobDispatcher for their job allocation, run the job, and submit the results back to the dispatcher before terminating.

Features

• A distributed functional map
• ProcessScheduler and GridEngineScheduler schedulers for testing tasks on a laptop, then scaling them up to the Grid Engine
• gridengine.Job API compatible with threading.Thread and multiprocessing.Process

Installation

Get gridengine from github and install using pip:

pip install git+https://github.com/hbristow/gridengine

This will automatically pull and build the dependencies.

Example

import gridengine

def f(x):
  """compute the square of a number"""
  return x*x

scheduler = gridengine.schedulers.best_available()

x = [1, 2, 3, 4, 5]
y = gridengine.map(f, x, scheduler=scheduler)

See gridengine/example.py for a runnable example.

Branches/Contributing

The gridengine branching schema is set up as follows:

master    reflects the current latest stable version
dev       the development track, sometimes unstable
issue-x   bugfix branches, with issue number from GitHub
feat-x    feature branches