-
Notifications
You must be signed in to change notification settings - Fork 46
Home
ROSS is an acronym for Rensselaer's Optimistic Simulation System. It is a parallel discrete-event simulator that executes on multiprocessor systems and/or supercomputers. ROSS is geared for running large-scale simulation models (millions of object models). The synchronization mechanism is based on Time Warp. Time Warp is an optimistic synchronization mechanism develop by Jefferson and Sowizral [10, 11] used in the parallelization of discrete-event simulation. The distributed simulator consists of a collection of logical processes or LPs, each modeling a distinct component of the system being modeled, e.g., a server in a queuing network. LPs communicate by exchanging timestamped event messages, e.g., denoting the arrival of a new job at that server.
The Time Warp mechanism uses a detection-and-recovery protocol to synchronize the computation. Any time an LP determines that it has processed events out of timestamp order, it "rolls back" those events, and re-executes them. For a detailed discussion of Time Warp as well as other parallel simulation protocols we refer the reader to [8].
ROSS was modeled after a Time Warp simulator called GTW or Georgia Tech Time Warp[7]. ROSS helped to demonstrate that Time Warp simulators can be run efficiently both in terms of speed and memory usage relative to a high-performance sequential simulator.
To achieve high parallel performance, ROSS uses a technique call Reverse Computation. Here, the roll back mechanism in the optimistic simulator is realized not by classic state-saving, but by literally allowing to the greatest possible extent events to be reverse. Thus, as models are developed for parallel execution, both the forward and reverse execution code must be written. Currently, both are done by hand. We are investigating automatic methods that are able to generate the reverse execution code using only the forward execution code as input. For more information on ROSS and Reverse Computation we refer the interested reader to [4, 5]. Both of these text are provided as additional reading in the ROSS distribution.