Distributed Shared Memory (DSM) provides an abstraction of shared memory on a network of workstations. Problems with existing DSM systems are lack of portability due to compiler and/or operating system modification requirements, and reduced performance due to significant synchronization and communication costs when compared to their message passing counterparts (e.g., PVM and MPI). Our approach was to introduce a new DSM consistency model, Lazy Data Merging (LDM), which extends Data Merging (DM). LDM is optimized for software runtime implementations and differs from DM by 'lazily' placing data updates across the communication network only when they are required. It is our belief that LDM can significantly reduce communication costs, particularly for applications that make extensive use of locks. We have completed the design of \"Eureka\", a prototype DSM system that provides a software implementation of the LDM consistency model. To ensure portability and efficiency we use only standard UniXTM system calls and a publicly available software thread package, Cthreads, from the University of Utah. Furthermore, we have implemented and tested some of Eureka's core components, specifically, the set of communication and hybrid (Invalidate/Update) coherence primitives, which are essential for follow on work in building the complete DSM system. The question of efficiency is still an open problem, because we did not compare Eureka with other DSM implementations.