Modeling Movable Components for Disruption Tolerant Mobile Service Execution

Abstract

Software as a Service relies on ubiquitous network access which cannot be assured in mobile scenarios, where varying link quality and user movement impair the always connected property. We approach this challenge by utilizing movable service components between a remote cluster, cloud, or server and the client device using the service. To overcome connection disruptions, service components are moved to the client prior to connection loss and executed locally. Although the basic concept is a brute force approach, challenges arise due to best fitting service decomposition, accurate estimation of connection losses, and best trade-off between moving service components and the overhead caused by this proactive fault tolerance mechanism.

This paper contributes to the general approach by presenting a system architecture based on an extended client/server model which allows to move components. Additionally, an analytical model is introduced for analyzing where to place service components best and extended to investigate failure rates and average execution time in different system configurations, i.e., different placement of service components either on the server cloud or client side. The models presented are based on Markov chains and allow to analytically evaluate the proposed system. Applied to a specific use case, we demonstrate and discuss the positive impact of placing components temporarily at the client in terms of failure rate and mean service execution time.