We introduce a class of networks called isotach networks designed to reduce the cost of concurrency control in asynchronous computations. Isotach networks support several properties important to the correct execution of parallel and distributed computations: atomicity, causal message delivery, sequential consistency, and memory coherence in systems in which shared data can replicate and migrate. They allow processes to execute atomic actions without locks and to pipeline memory accesses without sacrificing sequential consistency. Isotach networks can be implemented in a wide variety of configurations, including NUMA (non-uniform memory access) multiprocessors and distributed as well as parallel systems. Networks that implement isotach time systems are characterized not by their topology, but by the guarantees they make about the relative order in which messages appear to be delivered. These guarantees are expressed in logical time, not physical time. Physical time guarantees would be prohibitively expensive, whereas logical time guarantees can be enforced cheaply, using purely local knowledge, and yet are powerful enough to support efficient techniques for coordinating asynchronously executing processes. Empirical and analytic studies of isotach systems show that they outperform conventional systems under realistic workloads, in some cases by an order of magnitude or more.
All rights reserved (no additional license for public reuse)
Reynolds, Jr, Craig Williams, and Jr Wagner. "Isotach Networks." University of Virginia Dept. of Computer Science Tech Report (1995).
University of Virginia, Department of Computer Science