On the Evolution of Time Horizons in Parallel and Grid Simulations
arXiv:cond-mat/0401229 · doi:10.1103/PhysRevE.70.026703
Abstract
We analyze the evolution of the local simulation times (LST) in Parallel Discrete Event Simulations. The new ingredients introduced are i) we associate the LST with the nodes and not with the processing elements, and 2) we propose to minimize the exchange of information between different processing elements by freezing the LST on the boundaries between processing elements for some time of processing and then releasing them by a wide-stream memory exchange between processing elements. Highlights of our approach are i) it keeps the highest level of processor time utilization during the algorithm evolution, ii) it takes a reasonable time for the memory exchange excluding the time-consuming and complicated process of message exchange between processors, and iii) the communication between processors is decoupled from the calculations performed on a processor. The effectiveness of our algorithm grows with the number of nodes (or threads). This algorithm should be applicable for any parallel simulation with short-range interactions, including parallel or grid simulations of partial differential equations.
24 pages with 11 figures
References in corpus (8)
- Suppressing Roughness of Virtual Times in Parallel Discrete-Event Simulations
- Kolmogorov turbulence in a random-force-driven Burgers equation
- From Massively Parallel Algorithms and Fluctuating Time Horizons to Non-equilibrium Surface Growth
- Parallelization of a Dynamic Monte Carlo Algorithm: a Partially Rejection-Free Conservative Approach
- Soliton approach to the noisy Burgers equation: Steepest descent method
- Algorithmic scalability in globally constrained conservative parallel discrete event simulations of asynchronous systems
- Update statistics in conservative parallel discrete event simulations of asynchronous systems
- Discrete-Event Analytic Technique for Surface Growth Problems