The Impact of Lookahead on the Performance of Conservative Distributed Simulation

Website of Bruno R. Preiss

Next : The Role of Knowledge Up: Conference Papers Previous: Performance of Discrete Event

Bruno Richard Preiss and Wayne Mervin Loucks. In Proc. 1990 European Multiconference--Simulation Methodologies, Languages and Architectures , pages 204-209, Nuremberg, FRG, June 1990. Society for Computer Simulation[53].

It has long been recognized that the ability of a logical process (LP) to ``look ahead'' is crucial to the performance of distributed simulation using conservative synchronization. This paper examines previous definitions of lookahead and lookahead ratio--metrics for characterizing the lookahead ability of LPs.
These metrics do not reflect the use of null messages. In order to assess the impact of null messages on the performance of the simulation, we introduce the concept of null message lookahead and null message inverse lookahead ratio (NILAR). In this paper, we also propose a consistent set of definitions for lookahead and inverse lookahead ratio (ILAR) for customer-neutral systems.

The results discussed in this paper indicate that NILAR acts as a better predictor of simulation speedup than ILAR in the case of closed queueing networks with a large number of null messages. The results presented also indicate that conservatively synchronized distributed simulation implementations must have ILAR values close to one and NILAR values greater than one in order to achieve good speedup. However, ILAR values close to one and NILAR values greater than one do not necessarily result in good speedups. Other requirements are higher loads and limited fanout.

external Full text. Presentation slides. $external$ BibTeX entry.