HydraNet-FT: network support for dependable services

With the Internet increasingly being used as the access medium for a variety of critical services, there is a growing need to provide fault-tolerant (FT) services over internetworks, in a completely client-transparent fashion. We present HydraNet-FT, an infrastructure to dynamically replicate services across an internetwork and have the replicas provide a single fault-tolerant service access point to clients. HydraNet-FT uses the TCP communication protocol with a few modifications on the server side to allow one-to-many message delivery from a client to service replicas and many-to-one message delivery from the replicas to the client. A communication channel between the replicas provides atomicity and message ordering. A low-latency failure estimator is used to detect failures of servers in the system and to initiate fail-over mechanisms. An implementation and measurements on a local testbed show that the overhead of our scheme is reasonably small