Partitioned Global Address Space Languages
More than a decade ago, programmer productivity was identified as one of the main hurdles for future parallel systems. The so-called Partitioned Global Address Space (PGAS) languages try to improve productivity and explore a range of language design ideas. These PGAS languages are designed for large-scale high-performance parallel programming and provide the notion of a globally shared address space, while exposing the notion of explicit locality on the language level. Even so the main focus is high-performance computing, the language ideas are also relevant for the parallel and concurrent programming world in general.
As part of our research in the field of parallelism and concurrency, we studied the PGAS languages more closely to get a better understanding of the specific concepts they explore and to get a feeling for the tradeoffs of the various language design options. The result is a survey of the major PGAS languages, which was very recently accepted for publication in the ACM Computing Surveys.
The preprint of the paper is available as PDF and HTML version. The final edited version will probably take another eternity to appear, but oh well, that’s academia.
The Partitioned Global Address Space (PGAS) model is a parallel programming model that aims to improve programmer productivity while at the same time aiming for high performance. The main premise of PGAS is that a globally shared address space improves productivity, but that a distinction between local and remote data accesses is required to allow performance optimizations and to support scalability on large-scale parallel architectures. To this end, PGAS preserves the global address space while embracing awareness of non-uniform communication costs.
Today, about a dozen languages exist that adhere to the PGAS model. This survey proposes a definition and a taxonomy along four axes: how parallelism is introduced, how the address space is partitioned, how data is distributed among the partitions and finally how data is accessed across partitions. Our taxonomy reveals that today’s PGAS languages focus on distributing regular data and distinguish only between local and remote data access cost, whereas the distribution of irregular data and the adoption of richer data access cost models remain open challenges.