Client-transparent fault-tolerant Web service

Most of the existing fault tolerance schemes for Web servers detect server failure and route future client requests to backup servers. These techniques typically do not provide transparent handling of requests whose processing was in progress when the failure occurred. Thus, the system may fail to provide the user with confirmation for a requested transaction or clear indication that the transaction was not performed. We describe a client-transparent fault tolerance scheme for Web servers that ensures correct handling of requests in progress at the time of server failure. The scheme is based on a standby backup server and simple proxies. The error handling mechanisms of TCP are used to multicast requests to the primary and backup as well as to reliably deliver replies from a server that may fail while sending the reply. Our scheme does not involve OS kernel changes or use of user-level TCP implementations and requires minimal changes to the Web server software