Improving the design of a distributed, high performance and flexible simulator-to-GUI interface

摘要

We describe a flexible and high performance approach to separating and distributing a discrete event simulator's GUI and runtime results visualization from its associated simulator, so that the GUI/visualization can be run on a host that is physically separated from the host on which the simulation and all its proprietary data reside. While the concept is not new and many proprietary visualization toolkits allow socket and/or HLA/DIS access from a user's simulation, the approach described here lends much flexibility in linking simulators to GUI/visualizations. Specifically, our approach has several distinctive features including: 1) A VCR like control on the GUI/visualization side that can start, stop and pause/resume the remote simulation without getting into deadlock, 2) Autonomous operation of the simulator including support for bi-directional initiation of shutdown commands, and support for bi-directional transfer of required data files, 3) Support for a rich and user-extensible remote API that can be called from either side in a platform independent manner (without depending on strings-only commands), where arbitrary objects (which have encapsulated conversion to and from an extensible set of platform neutral transmission classes) can be sent as API parameters, and return values received in a blocking fashion without deadlock, 4) Support for the GUI/visualization to control multiple concurrently running remote simulation clients (after admitting/deleting them on the fly), 5) Support for the GUI/visualization to provide inter-client communication between remote simulations (the design uses additional threads with their own ports for pure messaging between these clients, whereas the display load of the GUI would slow down this messaging in a single-threaded design), 6) Publishing of TCP port(s) that can be read by a remote simulation that seeks to connect to the GUIl/visualization, and piggybacking of critical commands on return value of API calls (for speed, etc.). This approach and its implementation are general and object-oriented and can be used in any simulation-to-GUI/visualization context. In fact, with additional API calls to reach consensus and avoid deadlock, these techniques can be reused for the more difficult problems in distributed simulation design. The approach does not preclude the use of proprietary visualization toolkits that allow socket and/or FILA/DIS access. We can in fact design our GUI/visualization code so that it has all of the above listed features, and in addition, takes on the easier responsibility of displaying on the proprietary visualization toolkit. We describe our experience in incorporating this approach in our Joint MEASURE discrete event simulation framework, and outline the consequences and future direction of this work.