Evolution of the Milieu Approach for Software Development for the Polymorphous Computing Architecture Program.

摘要

A key goal of the DARPA Polymorphous Computing Architectures (PCA) program is to develop reactive closed-loop systems that are capable of being dynamically reconfigured in order to respond to changing mission scenarios. This is in contrast to current systems that are fixed in nature and rely on architecture and software optimizations targeted for specific missions. In order to accomplish this objective, a number of 'malleable' processing elements, runtime support software, compilers, and application development tools are being developed by a variety of research teams. Application software development for PCA systems is expected to be particularly challenging because of the need to respond to rapid changes in the mission requirements and environment. The authors of this talk have formed a team that is focusing its effort on developing application modeling tools and middleware libraries that assist in managing the complexity of developing. deploying. and maintaining PCA applications. The four high-level research objectives of this effort are as follows: (star) Study. prototype and develop a modeling language for streaming and threaded resources and components. (star) Perform Design Space exploration to enable PCA scheduling (including use of AI techniques). (star) Achieve System Synthesis and Generation (including performance monitoring and feedback) to support, and (star) Accomplish Dynamic Reconfiguration study and support for PCA.