Blast: Distributed Computing in Haskell

Blast is a pure Haskell library for doing distributed computing. It has the following characteristics:

• Works on any RPC backend. The current implementation runs on both local threads and CloudHaskell.
• Is based on 5 simple primitives, allowing the user to define his own syntax above them.
• Has a buit-in fail/safe mechanism in case of slave failure.
• Transparently works with both stateless and stateful slaves.
• Automatically handles slave caching.

Getting started

$ stack build

Builds the library and the examples.

Examples

Each example can be run on 2 backends:

• A local backend that uses local threads.
• A CloudHaskell backend.

To run an example on CloudHaskell:

• Edit Main.hs and choose the right CloudHaskell main function (the one that is suffixed with CH).
• Starts as many slaves as needed.

$ stack exec -- example-name slave host port

• Starts the master.

$ stack exec -- example-name master host port

I.E. The following commands starts the KMean example with 2 slaves.

$ stack exec -- kmeans slave localhost 5001
$ stack exec -- kmeans slave localhost 5002
$ stack exec -- kmeans master localhost 5000

Simple

A set a simple examples illustrating remote mapping and folding as well as an iterative distributed algorithm.

$ stack exec -- simple

KMeans

Implementation of the distributed KMean algorithm.

$ stack exec -- kmeans

WordCount

Counts the number of occurrences of each character in multiple files.

$ cd examples/WordCount
$ stack exec -- wordcount

License

Copyright (c) 2016-2017 Jean-Christophe Mincke.

All rights reserved.

Blast is free software, and may be redistributed under the terms specified in the LICENSE file.