Abstract

Modern computer systems are not only increasingly parallel, but also heterogeneous. Common examples include CPU -GPU combinations and IBM ’s Cell architecture. Placement, scheduling and indeed algorithm choices affect the overall execution time and, for portable programs, must adapt to the target machine at either load-time or run-time. We see these choices as preserving I/O determinism but exposing performance non-determinism.

We use Petri-nets as an intermediate representation for programs to give a unified view of performance non-determinism, including how to construct such nets given a model of a heterogeneous architecture. This includes some scenarios which other models cannot support.

Heuristics for finding optimal executions in these nets would not only solve the placement problem, but also programming model specific questions such as “which task should be given a lock first”.

This talk will be slightly less formal than normal, with more chance for discussion.