Performance Issues for Parallel Implementations of Bootstrap Simulation Algorithm

The solution of state-based stochastic models is usually a demanding application, then it is a natural subject to high performance techniques. We are particularly interested in the speedup of Bootstrap Simulation of structured Markovian models. This approach is a quite recent development in the perf...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2010 22nd International Symposium on Computer Architecture and High Performance Computing s. 167 - 174
Hlavní autoři: Czekster, R M, Fernandes, P, Sales, A, Webber, T
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.10.2010
Témata:
Analytical models
Automata
Computational modeling
Discrete-Event Simulation
Markov processes
Numerical models
Parallel Applications
Performance Evaluation
Stochastic Modeling
Storage area networks
Trajectory
ISBN:9781424482870, 1424482879
ISSN:1550-6533
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The solution of state-based stochastic models is usually a demanding application, then it is a natural subject to high performance techniques. We are particularly interested in the speedup of Bootstrap Simulation of structured Markovian models. This approach is a quite recent development in the performance evaluation area, and it brings a considerable improvement in the results accuracy, despite the intrinsic effect of randomness in simulation experiments. Unfortunately, Bootstrap Simulation has higher computational cost than other alternatives. We present experiments with different options to optimize the parallel solution of Bootstrap Simulation applied to three practical examples described in Stochastic Automata Networks (SAN) formalism. This paper contribution resides in the discussion of theoretical implementation issues, the obtained speedup and the actual processing and communication times for all experiments. Additionally, we also suggest future works to improve even more the proposed solution and we discuss some interesting insights for parallelization of similar applications.
ISBN:9781424482870
1424482879
ISSN:1550-6533
DOI:10.1109/SBAC-PAD.2010.28