A Comparative Study of Methods for Time-Deterministic Message Delivery in a Multiprocessor Architecture

In this paper we present a method, called TDC, for time-deterministic delivery of fine-grain messages in multiprocessor architectures with point-to-point interconnection networks. Other methods for scheduling predictable communication in such networks are flexible and efficient but suffer from a high cost when hardware-only implementations are required. Using results from a comparative study, we argue that efficiency in the scheduling technique can be traded for simplicity in the implementation. In particular, we show that if the simpler techniques of TDC are used instead of more complex methods, the implementation cost for communication mechanisms can be reduced by as much as 40% at the price of an overall performance degradation of no more than 15% in the application. In addition, the lower hardware requirements of TDC reduce the message processing delay by 33%.