PhD Research

QoS-Aware Service-Oriented Middleware for Pervasive Environments

 
My PhD thesis aims at designing and implementing a QoS-aware service-oriented middleware for pervasive environments. The main contributions of my PhD are: (1) a semantic end-to-end QoS model that enables shared understanding of QoS in pervasive environments, (2) an efficient QoS-aware service composition approach allowing to build service compositions that are able to full the user's functional and QoS requirements, and (3) a QoS-driven adaptation approach to cope with QoS fluctuations during the execution of service compositions. 
 
The mentioned contributions are implemented within a middleware platform called QASOM and their efficiency is validated based on experimental results. Below, I give an overview of each contribution.
 

A Semantic End-to-End QoS Model for Pervasive Environments

The first contribution of my thesis consists in defining a QoS model that provides the appropriate ground for QoS awareness in pervasive environments. Accordingly, I propose a semantic QoS model that addresses QoS on an end-to-end basis, by covering quality properties of the main QoS stakeholders in pervasive environments; in particular the model deals with: (i) network and hardware resources, (ii) application services and (iii) end-users. The proposed model is expressed in the OWL-DL ontology language1. Further details about this model are given in the following link:
 
 

A QoS-aware Service Composition Approach

 I introduce a QoS-aware service composition approach that enables building user tasks dynamically on-the-fly, while guaranteeing to meet associated QoS requirements. My approach is underpinned by a QoS-aware service selection algorithm (called QASSA) that addresses global QoS requirements (i.e., requirements imposed on the whole task), which is known to be a NP-hard problem. To solve this problem efficiently, QASSA introduces a novel heuristic based on data clustering techniques (namely the K-Means algorithm). The heuristic considerably reduces the number of investigated services by clustering them with respect to different QoS criteria, then performing the intersection of clusters. Thanks to this idea, QASSA achieves high timeliness and optimality. Additionally, it copes with major challenges of QoS-aware service selection in pervasive environments, notably:

  1. Considering end-to-end QoS requirements;
  2. Considering run-time QoS;
  3. Adaptation support; 
  4. Distributivity.

Further details about this model are given in the following link:

IEEE ICWS 2015_Nebil Ben Mabrouk.pdf (575,9 kB)

A QoS-driven Composition Adaptation Approach 

I introduce a QoS-driven composition adaptation approach that focuses on behavioural adaptation, i.e., finding an alternative behaviour to full the user task. An alternative behaviour can be defined as a composition of abstract services having a different structure (i.e., coordinated with respect to different composition patterns) or a different granularity (i.e., finer-grained or more coarse-grained activities). My adaptation approach introduces the concept Task Class, which denes functionally equivalent service compositions. Based on the task class concept, I reduce the behavioural adaptation to a graph comparison problem, specifically vertex disjoint Subgraph Homeomorphism (vdSH). vdSH determination is another important contribution of my thesis as it allows to establish the equivalence between complex service compositions and introduces a novel idea towards behavioural adaptation.
 
My PhD work is validated in the context of the SemEUsE and I-CROSS research projects (see the list of research projects). It was also the subject of a visiting PhD fellowship at the Honiden lab. (National Institute of Informatics, Tokyo, Japan).