Multicore vs Distributed Memory Performance Comparison

Title: Multicore vs Distributed Memory Performance Comparison

Proposer:

Phil Trinder
Phone: 0131 451 3435
E-mail: trinder@macs.hw.ac.uk
www:http://www.macs.hw.ac.uk/~trinder

Suggested supervisors:

Phil Trinder

Principal goal of the project:

To compare the parallel performance of shared-memory (multicore) and distributed-memory implementations of the same language.

Description:

Glasgow parallel Haskell (GpH) and SMP-GHC are two parallel implementations of the Haskell functional programming language. SMP-GHC is designed for shared memory machines, e.g. multicores. In contrast GpH is designed for distributed memory machines, e.g. clusters like the department's 32-Pentium Beowulf cluster. The project aims to compare the performance of the two implementations when executing a suite of 8 benchmark programs. If the comparison is done well the results should be publishable at a conference.

The objectives of the project are as follows.

To download SMP-GHC
To measure the parallel performance of the benchmark programs on the two or more architectures.
To write a publishable report on the results.
To assemble the parallel programs into a benchmark suite.

Resources required:

Linux,
Beowulf Multiprocessor
Glasgow parallel Haskell (GpH) (free to download)
SMP-GHC (free to download)

Degree of difficulty:

Medium

Background needed:

Familiarity with a functional language
A basic understanding of parallel computation

References:

[1] GHC Home Page

[2] GpH Home Page