Partition detection and optimistic commit for dynamically reconfigurable distributed databases

This paper addresses the problem of providing reliability in face of network partitioning for a dynamically reconfigurable distributed database system with an underlying hierarchical network topology. Solutions to the problems of partition detection and the proper termination of transactions are suggested. Passive partition detection protocols designed to handle node failures and connections are presented. The basis for these protocols is the idea of maintaining multiple network servers across the network to ensure communication within any cluster of two nodes. In the process, a network server allocation algorithm is derived to meet this need. This paper also proposes a new transaction management model based on the idea of non-blocking partition processing. An extended quorum-based commitment protocol and an optimistic quorum-based termination protocol are developed to support this goal.