Update Propagation in Bakunin Data Networks
In a Bakunin network data are replicated at all nodes in order to achieve very high data availability. Nodes operate autonomously, executing transactions even when they are cut off from the rest of the system. This means that transactions may read stale data. In spite of this, serializability can be guaranteed by placing restrictions on the types of transactions that a node can execute. These restrictions take the form of an acyclic Read-Access Graph. In addition, a special update propagation protocol is used to ensure that all nodes see data updates in the same order. In this paper we present several such
protocols. The protocols take advantage of the particular structure of the read-access graph to expedite propagation. We also define the notion of virtual serializability. It is a weaker form of serializability that allows speedier propagation.