Simplified synchronization through optimistic linearizability

We explore a novel programming model for multicore architectures in which shared data-structures are replicated per-thread. With a suitable mechanism for broadcasting data-structure operations (which may be wait-free or lock-free) there is no longer a need for elaborate, datastructure-specific synchronization (e.g. as seen in java.util.concurrent) at each replica. Moreover, threads can lazily apply the broadcast operations on an as-needed basis. In this position paper, we provide a formal semantics for this programming model but have not yet explored the practical viability of the approach. Nonetheless, previous work seems to indicate that replication, at least in the context of operating system internals, can yield performance improvement [1].