A Variation-Aware Quantitative Analysis And Optimization Framework For ThingML Designs

With the development of sensing,communication,and embedded computing tech-nology,the Internet of Things(IoT)has been widely adopted in a spectrum of areas.ThingML is a promising IoT modeling and specification language,which enables the fast development of resource-constrained IoT applications.Since these applications are deployed within open physical environments,their execution suffers from some uncer-tain factors,such as network delay.However,ThingML lacks the capability to model such uncertainties and quantify their effects.Therefore,within uncertain environments the performance and quality of generated IoT applications cannot be guaranteed.How to effectively evaluate the quality of application generated from ThingML within un-certain environment and obtain an optimal design become a major challenge.To solve these problems,this thesis makes three innovation as follows:1.We extend the syntax and semantics of ThingML modeling language,which en-ables the accurate modeling of performance variations caused by environments.This is because current version of ThingML assumes that IoT devices are deployed in an ideal environment,and the Quality of Service of generated application cannot be guaranteed.Consequently,some effective mechanism is needed in ThingML.2.We adopt the Network of Priced Timed Automate(NPTA)as the model of com-putation of our extended ThingML modeling language.Using our mapping rules,extended ThingML designs can be transformed into NPTA model for the quantita-tive analysis.By analyzing the performance of ThingML designs,IoT developers can make sure that the ThingML design achieve the desired performance require-ments within open physical environment.3.Based on the generated NPTA models,we propose a novel framework which en-ables the evaluation of Quality of Service(QoS),and adopt neural network to sup-port the quick exploration of system configuration with an optimal performance.This is due to the fact that if every configuration combination is verified indepen-dently,it will consume a large amount of time.The experimental results demonstrate that within uncertain environments,our pro-posed approach can effectively evaluate the QoS of ThingML-based application and find out some optimal system configurations to facilitate the decision making of ThingML developers.