Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Conservative parallel discrete-event simulation: Principles and practice

Thesis/Dissertation ·
OSTI ID:6038503
Simulation is one of the most important computational technologies in use today. Unfortunately, its importance is matched by its appetite for computational resources. These factors make parallel simulation a topic with far-reaching consequences in all fields of science and engineering. This thesis is concerned with one approach to this problem, conservative loose event-driven parallel simulation, the objective of which is to apply multiple processors to a single simulation run in an effort to reduce its time-to-completion. There are several factors that make parallel simulation difficult. First, the fact that a physical system has a high degree of concurrency does not necessarily mean that a simulation of that system will benefit from parallelism. The author introduces two simple analytic techniques that can be used to bound from above the speedup potential of parallel simulations. Second, a parallel simulation requires synchronization to ensure that the results obtained are equivalent to those of a sequential simulation of the problem. He argues that the availability of inexpensive, medium-scale and shared-memory multiprocessors mandates a re-examination of synchronization algorithms for conservative loose event-driven parallel simulation. His investigations lead to a novel synchronization technique called lazy blocking avoidance. His measurements show that lazy blocking avoidance performs at least as well as, and often substantially better than, two other synchronization methods that have been widely discussed in the literature (deadlock detection and recovery), and eager blocking avoidance.
Research Organization:
Washington Univ., Seattle, WA (USA)
OSTI ID:
6038503
Country of Publication:
United States
Language:
English