IRBP - A fault tolerant total order broadcast for large scale systems

Abstract

Fault tolerance is a key aspect on the development of distributed systems, but it is barely supported on large-scale systems due to the cost of traditional techniques. This paper revisits RBP, a Total Order Broadcast protocol known by its efficiency that presents some very interesting characteristics for scalable systems. However, we found a membership flaw on RBP that can lead to inconsistencies among correct processes. Hence, we propose iRBP, an improvement to the RBP algorithm that not only circumvents the membership weaknesses using recent membership techniques, but also improves its scalability aspects. © Springer-Verlag 2003.