Hobbes: Operating System and Runtime Research for Extreme Scale

PI Ron Brightwell, Sandia National Laboratories
Project Description

Four key challenges facing future large‐scale computing systems are; dramatically improve power efficiency, improve resilience in the presence of increasing faults, enable efficient data movement across deepening memory hierarchies and new storage technologies, and managing dramatically increased parallelism. To address these challenges, the operating system and runtime (OS/R) must take on more responsibility for managing more resources, like power and parallelism, and share more of the burden for insulating applications from the complexities of a system. Much of the focus of extreme‐scale system software in the last decade has been on measuring and characterizing the impact the OS can have on application scalability.

The Hobbes project is a collaboration of four national laboratories and eight universities with the goal of providing a system software environment that enables application composition through lightweight virtualization. Rather than providing a single unified OS/R that supports several parallel programming models, Hobbes is leveraging lightweight virtualization that provides the flexibility to construct and execute custom OS/R environments. While much of the exploration and development of the Hobbes software environment can be done at small scale, evaluating the scalability of OS/R interfaces and mechanisms at large scale is crucial. This project supports large scale testing and evaluation of OS/R implementation techniques to improve OS/R for future high performance computing machines.