Extant performance evaluation models for conventional multiprocessor systems have their limit to be directly applied to express what the dataflow computer architecture behaves. The limit came from the difference in computation models on which they are...
Extant performance evaluation models for conventional multiprocessor systems have their limit to be directly applied to express what the dataflow computer architecture behaves. The limit came from the difference in computation models on which they are based.
Even in the conventional queuing models for dataflow computers, ignoring important factors such as firing rules, organizations of processing elements, and thread scheduling overheads has limited their application in modelling the performance of dataflow computers.
A model proposed here, which has its basis on the dataflow semantics, could solve most of the problems appeared in the conventional models and had more expressive power than the previous ones.