A concurrent composite computational model for stochastic simulation
Abstract
The constructive processes required for performing stochastic simulation are usually not coherent and are divided into two independent phases: modeling (model construction) and simulation (model implementation and experimentation). The lack of coherence between both phases is often experienced as a nonlinear progression of effort during the employment of modeling and simulation. This disparity has long been a source of complaint from the modeler. It increases not only the lead time and complexity of modeling and simulation but also hampers use of the simulation technique. Since stochastic simulations tend to be compute-intensive applications, a significant fraction of the computational effort is spent in simulation calendar processing. Hence, choosing a good simulation calendar algorithm for simulation is a crucial issue. Past research has shown that simulation calendar processing can consume as much as 40% of the total simulation time. Therefore, the choice of a good simulation calendar algorithm can have significant impact on simulation time. There are two approaches to tackling this problem: (1) propose a new simulation calendar algorithm, (2) use the Standard Hold model to assess the performance of different simulation calendar implementations and choose the best. In this thesis, we will present an integrated scheme, named the $C\sp3$ (Concurrent Composite Computational) model, to tackle the problem of incoherence between the modeling phase and the simulation phase. The $C\sp3$ model can integrate the modeling phase and the simulation phase into a uniform whole and relieve modeling effort to some extent. It is based on the combination of event-scheduling, the process-interaction scheme and the object-oriented paradigm. It views the system as a collection of processes and objects and is able to replicate computations to perform distributed simulation in a heterogeneous environment. To demonstrate the utility of the proposed model, we also include examples of communication networks, computer systems, queueing network applications and a multidimensional integral computation. We will also present a new Markovian approach and new metrics to measure the dynamic performance of simulation calendar implementations. In contrast to the conventional static Standard Hold model, the new Markov Hold model is more general and can be used to examine how different implementations responded dynamically to dependent sequences of insertion and deletion requests. In this thesis, we present a systematic method for evaluating simulation calendars and the detailed mechanism of the $C\sp3$ model along with experiments on a networked computer environment.
Degree
Ph.D.
Advisors
Rego, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.