Hope: A Fault-Tolerant Distributed Pub/Sub Architecture for Large-Scale Dynamic Network Environment

摘要

Existing distributed XML-based pub-sub systems normally targeted on filtering precision, while lacked scalability and fault-tolerance concern. In a large-scale dynamic network, nodes may join/leave/fail at times, hence a scalable yet fault-tolerant architecture is crucial. In this paper, we proposed HOPE: a hypercube overlay based pub-sub architecture for efficient XML dissemination in large-scale dynamic network environment. We first proposed a highly space/time efficient matching scheme for representing and filtering XML document based on XPE efficiently, utilizing bloom filter data structure. We then developed a hypercube overlay, which exploits the Key-Based Routing (KBR) mechanism, to provide a fault-tolerant and scalable publication dissemination architecture. Extended simulation experiments show proposed system can scale well, and can tolerate moderate levels of nodes failure: it works successfully in 4096 nodes system that has 25% node's failure rate, with a replication factor of 8 (which caused link traffic overhead of only 500Byte/Sec).