M U S E

The Miami University Simulation Environment (MUSE) is a general purpose parallel simulation framework. It is has been designed to ease development (programming) of parallel simulations -- yes, it requires programming skills in order to be used. If you are planning to use one of the examples/applications then you need to directly work with those examples rather than with the MUSE kernel. On the other hand, if you are planning to build an application using MUSE or if you are performing parallel simulation research then you will need to work with the MUSE kernel.

MUSE is an acronym for Miami University Simulation Environment. It has been developed in C++ using object-oriented approaches and the Message Passing Interface (MPI) library. MUSE uses Time Warp and standard state saving approach to accomplish optimistic synchronization of the LPs. A conceptual overview of a parallel simulation is shown in the adjacent figure. A MUSE simulation is organized as a set of Logical Processes (LPs) that interact with each other by exchanging virtual timestamped events. An LP is implemented as a C++ class by overriding necessary methods in an muse::Agent base class. The input, output, and state queues used for rollback operations in Time Warp are managed by the muse::Agent base class in coordination with the simulation kernel.

The MUSE simulation kernel implements core functionality associated with LP registration, event processing, state saving, synchronization, and Global Virtual Time (GVT) based garbage collection. The kernel also handles the task of interacting with MPI for exchanging events between the MPI-processes. The kernel uses a centralized Least Timestamp First (LTSF) scheduler queue for managing pending events and scheduling event processing for local LPs. LPs are permitted to generate events only into the future -- that is, the timestamp on events must be greater than their Local Virtual Time (LVT). Consequently, with a centralized LTSF scheduler, event exchanges between local LPs cannot cause rollbacks. Only events received via MPI can cause rollbacks in MUSE. The scheduler is designed to permit different data structures to be used for managing pending events. MUSE includes several different schedulers and data structures to enable efficient parallel simulations.