MCFlow: A real-time multi-core aware middleware for dependent task graphs

Driven by the evolution of modern computer architectures from uni-processor to multi-core platforms, there is an increasing need to provide light-weight, efficient, and predictable support for fine-grained parallel and distributed execution of soft real-time tasks with end-to-end timing constraints, modeled as directed a cyclic graphs whose edges capture dependences among their subtasks. At the same time, there is a need to support state of the art programming models such as distributed components, whose ability to encapsulate functionality and allow context-specific optimizations is essential to manage the increasing complexity of modern distributed real-time and embedded systems and systems-of-systems. Real-time distributed middleware such as RT-CORBA has not kept pace with these developments, and a new generation of middleware is needed that can map these dependent subtask graphs onto distributed hosts with multi-core architectures, efficiently and within a simple, lightweight, and intuitive component programming model. To overcome these limitations, we have designed and implemented MC Flow, a novel distributed real-time component middleware for dependent subtask graphs running on multi-core platforms. MC Flow provides three new contributions to the state of the art in real-time component middleware: (1) a very lightweight component model that facilitates system integration and deployment through automatic code generation at compile time from a deployment plan specification, (2) transparent optimization of inter-component communication, and (3) the use of interface polymorphism to separate functional correctness from data copying and other performance constraints so that they can be configured and enforced independently but in a type-safe manner. Empirical evaluations of our approach in comparison to the widely used TAO real-time middleware show that MC Flow performs comparably to TAO when only one core is used and outperforms TAO when multiple cores are involved.