Lorenzo Donatiello - Alessandro Fabbri
Dipartimento di Scienze dell'Informazione
Università di Bologna
Mura Anteo Zamboni 7
I - 40127 Bologna (Italy)
email: {donat,fabbria}@cs.unibo.it
Distributed discrete event simulation is intended to achieve the highest speed-up with respect to the sequential execution of a simulation experiment by exploiting the concurrency of the events which occur in the simulated system. A distributed simulation metodology based on the concurrent construction of a distributed data structure representing the graph of causal dependencies between events, is discussed. Such a methodology uses a rollback mechanism to correct causal errors, and does not need any a priori decomposition of the simulator into processes since the whole active portion of the global state history is retained within the dependency graph. Single event elaborations (tasks) activate each other according to the event scheduling process, concurrently find the suitable state values each elaboration depends on, and opportunely alter the global state history. Each task is responsible of updating the dependency graph by inserting a node indicating the operations it performed, so that other tasks are able to detect any causal error due to incorrect utilizations of state values it might have done, and consequently activate rollback tasks to correct such errors.
We investigate coordination languages as support to the construction of distributed simulators based on the described methodology. In particular, we show that the abstraction of the tuple space computation model provided by coordination languages constitutes a suitable environment for the implementation of a distributed data structure representing the graph of causal dependencies between events.
Back to Fabbri's home
page