The DEVS environment for high-performance modeling and simulation

Abstract

S imulation is an ideal way to predict behavior in complex systems that we cannot otherwise test. For example, imagine trying to predict the behavior of a weapons system or a set of flood warning sensors by testing the real thing. Simulation offers definite advantages-safety and controlled circumstances, for instance-over the limited testing we could perform with an actual wartime battle or a real flood. The behavioral complexity that such large-scale modern systems can exhibit, however, demands computing power that exceeds current desktop technology. To meet such a challenge, we need high-resolution, large-scale representations of systems composed of both naturally occurring and manmade elements.In this article we describe a high-performance simulation environment we developed to model complex systems. We also discuss results we obtained by modeling and simulating a watershed. Environment overviewThe simulation environment we developed is called DEVS-C++. It is based on a modeling formalism called DEVS, which stands for Discrete-Event System Specification, as we will explain.We can characterize a high-performance, simulationbased design environment with two levels of intensive information processing:♦ At the decision-making level, the environment searches many alternative designs and associated models.♦ At the execution level, simulations generate and evaluate complex candidate model behaviors, possibly interacting with human participants in real time.Our simulation environment has four layers: simulation (lowest), modeling, optimization, and decision making (highest). Each builds on lower layers to implement its functionality.As Figure 1 shows, processes execute concurrently in a heterogeneous, distributed computing environment. 1 We use genetic algorithms in the optimization layer to search through a model space to find models that will effectively predict system behavior.The functions of a GA are distributed among software components called agents. Each GA agent has access to a simulator for executing experiments. Experiments consist of several trials, testing how well a particular structure works in a problem environment. We represent the environment as a simulation model, controlled and ♦ ♦ ♦ DEVS-C++, a high-performance environment for modeling large-scale systems at high resolution, uses a discrete-event formalism called DEVS to represent both continuous and discrete processes. A prototype suggests that the DEVS formalism can combine with genetic algorithms running in parallel to serve as the basis of a very general, very fast class of simulation environments.♦ .

DEVS modeling of mobile wireless ad hoc networks

“…29 and 30 also shows that the node located at the right most end of the hop-count plane (i.e. located at cell (12,18)), can reach both gateways. This node chooses the path with the lowest hop count and sets its value accordingly (i.e., 30 instead of 50).…”

“…31 with hopcount value of 60 moves to a new cell (18,9) in Fig. 30, the new cell is not aware of the presence of the node in its neighborhood with a hop-count value of 10.…”

“…Nine different collision scenarios are created from this initial topology. Four of these collisions are between static and mobile nodes (potential collisions between node pairs located at cells: (0, 0)-(1, 1), (1, 4)-(0, 5), (9, 10)- (10,11), (18,12)- (19,13)), three of them are between two mobile nodes (potential collisions between node pairs located at cells: (2, 12)- (4,14), (6, 1)-(8, 1), (12,3)- (12,5)) and two of them are between a mobile node and a gateway (potential collisions between node pairs located at cells (14,18)- (15,19), (16,18)- (15,19)). Fig.…”

“…The toolkit has been built as a set of independent software pieces, each of them independent of the operating environment chosen. Our goal was to build models of ad hoc networks using the advantages of DEVS and Cell-DEVS for the following reasons: efficiency (by describing a high level specification of the problem to be modeled, we have reduced the effort needed in developing the application) and high performance (the models execute using a discrete-event approach, which, as shown in [18][19][20], provide higher precision and speedups than the discrete time approaches). In [18], the authors showed that the discrete-event nature of DEVS models combined with parallel simulation techniques can produce speedups of up to 1000 times.…”

“…• The potential to automatically deploy models that have been tested on the simulation environment into the actual networking hardware, converting them into the real applications [17,18].…”