This paper proposes a task graph (TG) model to represent a singlediscrete step of a computational fluid dynamics (CFD) application. TheTG model is then used to predict the performance of a set of taskassignment heuristics developed based on the constraints inherent in theCFD problem. Two primitive assignments, the Largest Task First and theMinimum Task First, are systematically enhanced by the integration ofthe status of the processing units and communication costs. Evaluationis performed on a synthetic TG with tasks defined in terms of the numberof gridpoints in predetermined grid zones. A realistic problem witheight million gridpoints is also used as a test case
展开▼