Retargeting Embedded Software Stacks for Many-Core Systems

As recent technology trends usher us into the many-core era, software applications must scale as 1000s core become available in a single chip. Middleware is a key infrastructure component between applications and the operating system. Therefore, new middleware mechanisms must be developed to handle scheduling, resource sharing, and communication on platforms with 100s and 1000s of cores. The solution must help application developers create concurrent software and must be easy to use. Real-Time Innovations (RTI) and the University of North Carolina (UNC) Real-Time Systems Group have teamed up to create mechanisms for scalable, high-performance, and adaptable scheduling and communication for many-core systems. Our solution has four key innovations: a component framework to facilitate higher concurrency in real-time application integration, a smart scheduler to ensure little capacity loss of the processor, improved RTI DDS messaging middleware using concurrency patterns to achieve higher throughput, and a middleware transport optimized for sending messages across 1000s of cores in a closely distributed and potentially heterogeneous environment. We validated our approach using prototype implementations and through experimentation. Our results indicate that, when commercialized, our approach will help existing as well as new applications to scale to 1000s of cores in a cost-effective way.