End-to-End Reliability for Best-Effort Content-Based Publish/Subscribe Networks

摘要

When it conies to reliability, there are two main categories of distributed publish/subscribe systems: reliable systems and best-effort systems. The former gives the highest priority to guaranteed and ordered delivery while the latter aims for high throughput and low delays. We propose a method to improve the delivery guarantees of the basic unreliable service offered by a best-effort publish/subscribe system. This method does not require any modification to the system's protocols or broker software, and instead simply uses the system's publish/subscribe API. The method is based on a technique, similar to reliable multicast, that enables subscribers to cooperatively recover lost messages. We experimentally demonstrate the effectiveness and performance of our recovery scheme in the presence of frequent message losses, and show that it enables subscribers to recover more than 70% of lost messages with minimum negative effects on