View on GitHub

DES Metrics

Profiling discrete-event simulation models

Download this project as a .zip file Download this project as a tar.gz file

DES Metrics: Quantitative Metrics from Discrete Event Simulation Models

Overview

This project is designed to capture run time metrics of Discrete Event Simulation (DES) models in order to evaluate quantitatively the event dependencies found therein. While the initial goals are to learn more about the structure and nature of the simulation models and their suitability for parallelism. Opportunities for learning about the simulation models in order to draw any meaningful conclusions about them will also be incorporated. For example, anything we can learn about relationships between simulation objects that could be used to derive strategies for static analysis and model transformation/optimization by mechanized tools would also be a highly desirable objective. Ideally the approach will capture the profile data in a generic intermediate representation so that the captured data can analyzed from a large number of DES simulation tools. While initially we will develop tools and methods to study individual simulation models, the ultimate goal will be to figure out how to bring information together from multiple simulation models to draw conclusions on the common characteristics of, in particular large, simulation models. Initially we will focus on the capture and analysis of event data.

This project is developed from a PDES centric view. In particular, we use term "simulation object" throughout which is not necessarily a strict construct within a sequential simulator. However, we will stick with this it will help us to effectively organize and structure our profile data and analysis activities in ways that will be most useful for us and our parallel simulation work.

Strategy

The general strategy will be to instrument existing simulators to capture event data for offline analysis. Ideally we will use a wide array of different discrete event simulators containing simulation models written by others (and ideally, domain experts from across multiple disciplines). This will require that we have source code to the simulator or programming cooperation from the community of the target simulation engine.

The second (larger) part of this project is the data analysis activity. The objective of this is to discover meaningful understandings of the characteristics of the simulation models. While there are specific analysis results that we expect to see in this phase, as we advance our studies in this space, we are likely to expand the types of analyses that performed.