Rainbow OS: A distributed STM for in-memory data clusters

Making parallel programming safe and intuitive still remains a major challenge in view of the fact that the exploitation of parallelism is not only desirable for multicore architectures but also for distributed systems such as clusters and grids. We claim that Software Transactional Memory (STM) is one of the more promising approaches to simplify the parallel programmer's job. Our Rainbow OS offers a streamlined in-memory data facility, accommodating many distributed application scenarios using a lean and teachable STM approach. The paper gives an overview and then focuses on memory management strategies and how Rainbow OS can host multiple, coexisting, custom-tailored memory consistency models within its transactional environment. We explain the benefits of our cluster-wide STM for distributed programming, for consistent memory image checkpointing and for system-level recompilation with on-the-fly code replacement. We also describe how the transactional programming paradigm can be used inside the operating system, particularly for device driver development.