ParaPLoP'09: First Workshop on Parallel Programming Patterns

I'll be attending ParaPLoP 2009 over the next few days. It's a patterns workshop along the traditions of PLoP but focusing exclusively on patterns for parallel programming.

I am the author and co-author of the following three patterns paper that you can find on the program page:

• Barrier Synchronization
• Patterns for Collective Communication
• Patterns for Topology Aware Mapping

Most of the patterns submitted are part of the Our Pattern Language (OPL): A Design Pattern Language for Engineering (Parallel) Software catalog proposed by Kurt Keutzer (EECS UC Berkeley) and Tim Mattson (Intel). OPL is based on Tim's earlier book Patterns for Parallel Programming.

While we have already identified quite a few patterns, there's definitely still a lot of work that needs to be done to produce a sizable pattern language that covers most of the patterns that most programmers will encounter in their programming career. However, I think that the layered approach that the Berkeley folks are advocating has great potential. The layered approach allows programmers to focus on different patterns depending on their skill sets and contributions to their project. At the top layer, application programmers focus on understanding the high level patterns and take advantage of the parallelism by using libraries and frameworks. And at the bottom layer, platform programmers focus more on understanding the low level patterns and develop libraries and frameworks to be used by other programmers.

Writing good software is hard. And writing good parallel programs is even harder. Patterns help make the task easier by showing the best practice principles that novice parallel programmers can learn from. And it's great that both UIUC and UC Berkeley are working together to catalog such patterns.

Ralph Johnson is leading the patterns project as part of the UPCRC at UIUC. And one of his projects is to mine for patterns by examining how different algorithms are expressed in different languages and frameworks, in particular those developed at UIUC (e.g. Deterministic Parallel Java, Charm++, Actor Foundry). Kent Beck, Samira Tasharofi, Amin Shali and I will be contributing to this project and our preliminary results can be found here.