A Framework for Adaptable Operating and Runtime Systems

Modern ultrascale machines have a diverse range of usage models, programming models, architectures, and shared services that place a wide range of demands on their operating and runtime systems. Full-featured operating systems can support these requirements, but the generality comes at a cost in performance. As we move toward petascale systems with hundreds of thousands of processors and multiple levels of parallelism, the relative cost of this overhead will increase. Lightweight operating systems, in contrast, provide efficient access to underlying hardware, but they offer only a limited set of system services and are not easily ported. We propose to offer both efficiency, extensibility, and portability by developing a framework for configuring and building operating and runtime systems using composable micro-services that can be combined to provide only the features needed for a given ultrascale environment.

This framework explores several important areas that have been identified as critical for achieving maximum effectiveness and efficiency for application developers and users for next-generation systems. In particular, our framework is designed to support the identification and development of micro-services specifically for the architectural innovations of next-generation hardware. In order to more easily adapt to the changing needs of applications, our framework is designed to support multiple management policies, with a focus on the scalability and usability requirements of high-performance computing. Finally, our approach provides a flexible mechanism for exploring and accomodating fundamental requirements of operating and runtime systems on ultrascale systems so that performance and scalability can be achieved and maintained.