Workshop: High Performance Computing for Research - 5 May 2014, Dunedin

9:30-10:45am

This session will introduce the audience to some concepts used in HPC and scientific programming in general.

We will start by looking at ways to split problems into manageable chunks that can be processed in parallel, known as problem decomposition. The session will be rich with examples, from bioinformatics to physics.

Further on, more terminology will be introduced. The audience will be exposed to the concepts of task and data parallelism. We will also look at what it means for something to be "embarrassingly parallel" or "tightly coupled".

As we near the end, we will cover some of the issues related to dealing with memory in an HPC system. In particular, we'll split the problem between sharing memory in a single node and working with memory across nodes in distributed system. This will bring us towards OpenMP and MPI, which are explained in depth later in the day.

10:45-11:15am

Morning Tea

Parallel Programming - OpenMP
Gene Soudlenkov

OpenMP is a technique for sharing memory between processes within a single hardware node. What is interesting about the technique is that legacy code that uses tight loops can be sped up with simple compiler directives.

As well as providing a theoretical overview, Gene will be running code directly on NeSI’s facilities.

12:30-1:30pm

Lunch

Using NeSI's HPC Platforms
Peter Maxwell & Ben Roberts

Learn about each of NeSI's HPC facilities and how to submit work to them. We will be covering a number of different techniques. Each has its own trade off between simplicity and customisability.

Afternoon Tea

MPI Programming with Python
Sung Bae & Jordi Blasco

The Message Passing Interface (MPI) is the de facto standard for parallel computing on high performance computing clusters.  Sung and Jordi will cover the primitives offered by MPI and how they can be combined to perform meaningful parallel computations.

Download: Source files (.tgz)