Abstract. This paper proposes a novel scheme, named ER-TCP, which transparently masks the failures happened on the server nodes in a cluster from clients at TCP connection granularity. Connections at the server side are actively and fully replicated to remain consistency so as to fail themselves over healthy parts during failure. A log mechanism is designed to cooperate with the replication to achieve small sacrifice on the performance of communication and makes the scheme scales beyond a few nodes, even when they have different processing capacities. The scheme is justified by experiments conducted on prototype implementation.