A Generic Approach to Dependability in Overlay Networks

Overlay networks are virtual communication structures that are logically “laid over” underlying hosting networks such as the Internet. They are implemented by deploying application-level topology maintenance and routing functionality at strategic places in the hosting network [1, 2]. In terms of dependability, most overlays offer proprietary “self-repair” functionality to recover from situations in which their nodes crash or are unexpectedly deleted. This functionality is typically orthogonal to the purpose of the overlay, and a systematic and complete approach to dependability is rarely taken because it is not the focus of the work. We therefore propose to offer dependability as a service to any overlay. Dependability is a well-studied field in distributed applications, but many of the existing approaches for applications are unsuitable for overlays; a common approach is to submit an application to a fault-tolerant framework for controlled execution within that framework [3]. We suggest instead that a dependability service for overlays is built only from decentralized, lightweight agents which exist alongside overlay nodes, operating at the same level and with the same resources available to their nodes. In addition, such agents should maintain only soft state which can be re-built automatically simply by existing in the environment, making the service inherently self-repairing. An architectural model of this is shown in figure 1.