Gossip-based Causal Order Delivery Protocol Respecting Deadline Constraints in Publish/Subscribe Systems

Abstract

Publish/subscribe systems based on gossip protocols are elastically to scale in and out and provides suitable consistency guarantees for data safety and high availability but, does not deal with end-to-end message delay and message order-based consistency. Especially in real-time collaborative applications, it is possible for the messages to take each a different time to arrive at end users. So, these applications should be based on P/S infrastructure including dealing with message deadlines and message ordering consistencies. Gossip communication is becoming one of the promising solutions for addressing P/S scalability problems in providing information propagation functionality by exploiting a mixture of diverse consistency options. In this paper, we present a new causal order protocol based on scalable P/S architecture for real-time collaborative applications in social web platforms to guarantee causally ordered message delivery, respecting deadline-constraints from brokers to subscribers. In the proposed protocol, every broker manages a 2-dimensional vector, representing its knowledge of the last message sent by each broker at a certain time. But, every broker disseminates a multicast message with a 1-dimensional vector, the time-stamped information that represents the maximum number of gossip rounds to subscribers because all messages disseminated by brokers have the same deadline as the maximum number of gossip rounds. Therefore, the proposed protocol for P/S based on gossiping results in very low communication overhead from brokers to subscribers in the context of respecting deadline-constraints.