Software Partitioning for Distributed, Sequential, Pipelined Applications

An examination of real-time performance analysis and allocation of system resources to meet throughput and response time requirements has aided in the development of a methodology for structuring software for real-time distributed systems. A queuing network model is used to estimate real-time performance for alternative software partitions. Those partitions that can satisfy user performance requirements while requiring minimal resources are identified. A 2-queue model, representing module service and communications, is evaluated using queuing network approximation. The results demonstrate that an optimal solution set can be determined with a branch and bound approach. The initial version is being improved to permit an expansion of its capability to allow a user to interactively explore performance variations for a selected structure as parameter values vary.