Skip to Main Content
Writing software for distributed systems is a complex task and gets even harder when shared data is replicated among nodes. Transactional memory is a promising technology for dealing with both synchronization and data consistency issues. Rainbow OS, a distributed operating system for PC clusters, employs this concept in a distributed fashion providing a cluster-wide transactional distributed memory (TDM). In the past Rainbow OS could only use a single processor in each cluster node. In our paper we present a lean way to integrate multiprocessor support into an uniprocessor TDM system, without sacrificing the programming model or other crucial design aspects. We discuss architectural considerations and their influences on the TDM subsystem.