δ-Fault-Tolerant Publish/Subscribe Systems

In this paper, we study reliable distributed publish/subscribe (P/S) systems that can “tolerate” mul-tiple simultaneous node crash failures. We formally define a routing consistency property, and proposescalable algorithms that establish and maintain consistency in order to guarantee reliable, in-order, andduplicate-free delivery of messages. Furthermore, we introduce a system configuration parameter, δ,that corresponds to the maximum number of simultaneous node failures that do not compromise P/S re-liability guarantees or prevent system operations. This is achieved via replication of routing informationin a fully decentralized manner compatible with the multicast nature of distributed P/S systems. More-over, we assume node failures are transient and devise algorithms that allow failed nodes to “recover”by synchronizing with other nodes. A recovered node acts as if it has never failed before, and can fullyparticipate in message forwarding.