Browsing by Author "Hendijani Fard, Fatemeh"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Detecting and Fixing Emergent Behaviors in Distributed Software Systems Using a Message Content Independent Method
    (2016) Hendijani Fard, Fatemeh; Far, Behrouz; Krishnamurthy, Diwakar; Moussavi, Mahmood; Helaoui, Mohamed; Bonakdarpour, Borzoo
    Distributed software Systems (DSS) and Multiagent Systems (MAS) as a sub-class of DSS can provide efficient and cost effective solutions for a wide range of applications. The distributed functionality and/or control in these systems and the local view of the scenarios of the systems can lead to unexpected behavior during execution time, known as Emergent Behaviors (EB) and Implied Scenarios (IS), which was not evident in the requirements and design phase. The new scenarios that are implied to the system can degrade the quality of service and/or cause irreparable damage. Detecting and fixing EB/IS in the early phases, may save costs of software projects by a factor of 20 to 100. In this thesis, we are investigating a new methodology for modeling and analyzing the behavior of software components/agents in order to certify their behavior in advance. Our research questions are: Q1: Is there any methodology that can detect common EB/IS in DSS/MAS without modeling the internal information/knowledge used in software components/agents? Q2: Is there a general approach that can detect EB/IS without human interference and is fully automated? First, we devised a catalogue of the common EB/IS that can arise in DSS/MAS. One of the main advantages of this catalogue is categorizing the EB/IS based on the reasons of occurrence, which helps in devising specific algorithms to detect each type of EB/IS, and can lead to devising solution repositories. The other contribution of our work is devising new modeling based on state machines and social network analysis. This modeling is a general method and can be implemented fully automated. Also, we devised algorithms for detecting the agents that will not show EB/IS in the system as a pre-processing phase. For classes of EB/IS in the catalogue, the detection methodology is devised and recommendations on how to fix the problem are provided. The results of our work shows that all of the EB/IS in various case studies specified in the literature can be detected with our method. Moreover, a new EB/IS is introduced which only can be detected with our modeling.