An architecture for task execution in adverse environments

Abstract

This thesis describes an architecture for distributed computation in mobile environments. Here, the workflow stands for the operational aspect of a work procedure. This notion subsumes the structure of the tasks, who performs them, what their operation structure is, how they are synchronized, how information flows to support the tasks and how they are tracked. The designed architecture is tested in a proof-of-concept implementation named Distributed Workflow Execution Architecture for Mobile (DWEAM). The interest in distributed workflow execution architectures is long-standing. However, there has not yet been an adequate treatment of workflow execution in mobile systems. The benefits of such a system would be collected by users who perform a coordinated task in a complex environment and must establish own coordination infrastructure to do so. Typical users are the members of emergency rescue teams, i.e.~police, fire brigade or medical personnel, when handling an incident, where the cooperation between the team members is hindered by mobility and adverse communication conditions. The communication via the gsm networks, that are used in small scale operations in urban areas cannot offer appropriate quality of service in face of escalation or infrastructure damage. The now-aged broadcast radio (i.e. walkie-talkie) typically does not support the coupling with information systems, as it is is intended for the communication between humans. We therefore consider a system architecture that addresses the communication issues, while being infrastructure-independent and supporting both the work procedures for human operators, as well as that of the information systems. First, the DWEAM problem is put into a broader context of mixed-initiative multi-actor, multi-agent systems. It is seen that DWEAM is but a single component of a larger system, called combined. The unifying description of combined is given, and the requirements for the integrated combined system are given. Following this description, we give an overview and commentary of the related work. We describe the toolkit that is used in the subsequent chapters, which relies on the usage of the Object-Z language, and the process model of the cpn. We then formalize the task of DWEAM, and complete the description of the used framework by specifying the distributed blackboard. We then describe the method used to cast the informal descriptions of inter-related tasks into a cpn. The cpn description is then converted into an implementation using a distributed blackboard. Afterwards the description of the operating environment and the storage model is given. We investigate the connectivity function for a set of nodes in two dimensions. Having done that, we turn our attention to the storage model, where we estimate the performance of the data partitioning to achieve the token preservation. For data distribution, the notions of producers and consumers are introduced, and the Service Discovery Problem (SDP) is defined. The Core-Based Tree (CBT) is afterwards used to provide a solution to the SDP, by finding the producers and consumers which are compatible, i.e. those that communicate through the delivery of data objects. A detailed analysis of the CBT construction algorithm is given, with the cpn descriptions of its phases and the discussion about its performance. All the data distributed in the DWEAM system must conform to the Dataspace model. It then becomes possible to define the matching algorithm. In it, the CBT structure constructed earlier is used to compute the matching between the compatible producers and consumers. The proof of the matching algorithm is given, followed by the cpn description of the implementation. The thesis is concluded by the list of contributions, the explanation of the outlook of distributed workflow execution in the contemporary context, and gives pointers to future work.