90 likes | 227 Views
Toward a Modeling Theory for Predictable Complex Software Designs. by Levent Yilmaz Auburn Modeling and Simulation Laboratory Department of Computer Science & Engineering Auburn University Auburn, AL 36849. Predictability: Toward a Modeling Theory for Predictable Complex System Designs
E N D
Toward a Modeling Theory for Predictable Complex Software Designs by Levent Yilmaz Auburn Modeling and Simulation Laboratory Department of Computer Science & Engineering Auburn University Auburn, AL 36849
Predictability: Toward a Modeling Theory for Predictable Complex System Designs • Off-Line Prediction with Model-Based Reasoning • On-Line Prediction: Exploratory Multisimulation of Alternative Designs with On-line Model Recommenders
Predicting the Behavior of “Possible” Systems using Simulation Modeling • Motivation:Large complex systems constantly evolve and their maintenance involves predicting and reasoning about the behavioral impacts of changes (i.e., substitutions) and alternative design organizations. • The Need: Need to compare design alternatives by analyzing the options of many possible systems, for which simulations may or may not yet have been built; hence may not be empirically studied. • Problem: Can simulation modeling excite an emerging prospect for developing a software design theory that supports off-line, as well as on-line exploration and reasoning about the behavior of “possible” systems? • Proposed Strategy:A hybrid approach that combines model-based reasoning with dynamic exploratory multisimulations to facilitate “effective” analysis for predictable design evolution.
Off-Line Prediction: Modular Reasoning about Complex Software-Intensive Systems • Designing systems as assemblies of standard component models whose behaviors are already understood in isolation. • System’s behavior must be understood using the component models and their localinterconnections. • Approach: • Describe the behavior of each simulation model component in isolation using a mathematical model. • Use the connections among components that depict the composition style and mathematical models of model components to derive a mathematical model of the behavior of the system as a whole. • Enforce the property that mathematical model of the behavior of the system involves modular composition of the mathematical models of components.
Component-Connector Approaches • At best an ad-hoc, non-uniform framework for composition rule and behavioral reasoning. • Computed properties are features of the global system. • No leverage for “effective” modular reasoning.
A Different Taxonomy: Reexamining the System Entity-Structure/MB Framework P1 P1 P1 P1 P1 B B P2 P2 P1 P1 P1 A A A A A B A A A A B B • Develop a mathematical model that captures system properties of interest. • Determine components (operands) and operators. • Syntax rule: Bind actual parameters to formal parameters. • Behavioral rule: Templates are functions applied to mathematical model of components.
Parametric Simulation with DEVS Templates Abstract Template Simulation Templates S S Abstract Component S S simulates S1 S B simulates L L S L1 L L2 A A A B simulates A1 A2 A3 B1 B2 Concrete components
On-Line Prediction: Multisimulation of Alternative Design Configurations with Model Recommenders • Multisimulation is dynamic, exploratory, and simultaneous experimentation with alternative design configurations. • Given that complex, dynamically reconfigurable, distributed systems that operate in unpredictable context are common in today’s component-based mission-critical systems, • Multisimulation with model recommenders can be useful in at least two ways: • Design Time: Simultaneous run-time exploration of the utility and effectiveness of alternative design decisions (i.e., configurations) under emerging conditions. • After Deployment: Symbiotic simulation to support making recommendations on possible system configurations that help achieve or optimize quality objectives of interest.