Design and Implementation of a Fault Tolerant Single IP Address Cluster

An F-FTCS mechanism that develops a fault tolerant single IP address cluster for TCP applications is proposed. The FTCS mechanism performs fine grain load balancing by handling all incoming TCP connection requests with a master node. Three fail-over algorithms are designed and implemented to carry out the fault tolerant FTCS mechanism. Discarding and Gathering Algorithms discard and gather TCP connections whose state is SYN-RECEIVED, respectively, at failure. A Scattering Algorithm synchronizes the information between nodes in the failure-free phase. These three algorithms are evaluated on Core 2 Duo machines. The Discarding Algorithm recovers from a failure from 440 to 950 msec earlier than the Gathering Algorithm, but it requires reprocessing the discarded TCP connection requests. The Scattering Algorithm requires from 120 to 160 usec more overhead during processing of a TCP connection request than that of the original FTCS mechanism.