 |
The use of high performance computers has been steadily increasing in recent years, from the core scientific community to a wider range of users. However, there is a cost in migrating to parallel systems in the effort required to port application code. The selection of the hardware platforms and of the parallelisation methods most appropriate, are important decisions that influence the resulting performance. The achievement of performance is of key interest. Studies can be formulated to produce performance predictions which can aid this decision process. However, to make these formulations accessible, a performance tool has been developed and is described here. It consists of a language that can express performance issues from a user's perspective, a preprocessor for conversion into Mathematica, and an evaluation and a run-time package for producing and visualizing performance predictions.
|
|
For the newest resources, visit Wolfram Repositories and Archives »
