Distributed Overlay Maintenance with Application to Data Consistency

Abstract

The proliferation of connected device rises new challenges for data maintenance in terms of efficiency, scalability and security. In the context of data replication at a large scale, weak consistency has been adopted as a standard in uniformly trusted systems. Recently, a distributed framework has been proposed to handle untrusted systems, by adding access control at the file level. Yet, if no particular care is paid to the connectivity of the network used for replication, w.r.t. dynamically evolving access rights, updates at replicas may not propagate correctly.

This paper proposes the design of a middleware application in charge of maintaining the connectivity of the overlay used for update propagation. We propose to improve the resilience of any overlay that may be disconnected as access right policies are modified or as critical nodes crash. This is achieved by reacting to a blocking node by rewiring the overlay in order to preserve at least one path between any pairs of replicas, while conserving the structural characteristics of the overlay. Simulations show that at a tunable overhead, connectivity is maintained, despite local decisions that would have otherwise disrupted the replication service.