Abstract
With the popularity of the Internet, many popular web sites are heavily loaded with huge numbers of requests. A scalable clustered server can be built to improve the performance of the web sites incrementally.However, as more and more web clients demand not only performance but also reliability and availability,designing a highly available cluster system becomes a very important topic.In this thesis, we focus on an important technique in high-availability cluster systems, the it heartbeat service.Traditional heartbeat systems mainly follow a client/server model, in which a monitoring node sends heartbeat messages to detect the liveliness of the monitored nodes.We propose here the it multinode heartbeat, which provides heartbeat service in a cluster system in which the nodes are peers. Our system allows multiple nodes to fail while still maintain proper monitoring relationship among the nodes.We have implemented our multinode heartbeat prototype, and we demonstrate its robustness by analyzing the system from the perspectives of failure detection, reconfiguration, and recovery from failure.