The Wackamole Approach to Fault Tolerant Networks Demo

Maintaining the availability of critical network servers is an important concern for many organizations. Server redundancy is the traditional approach to provide availability in the presence of failures. From the client perspective, a network-accessible service is resolved via a set of public IP addresses specified for this service. Therefore, the continued availability of a service via these IP addresses is a prerequisite for providing uninterrupted service to the client. In order to function correctly, each of the service’s public IP addresses has to be covered by exactly one physical server at any given time. If no physical server covers a public IP address, the clients will not receive any service. On the other hand, if more than one physical server is covering the same IP address at the same time, the network might not function properly and clients may not be served correctly. A sizable market exists for hardware solutions that maintain the availability of IP addresses, usually via a gateway that hides the actual servers behind a smart switch or router in a centralized manner. We present Wackamole [1], a high availability tool for clusters of servers. Wackamole ensures that a server handles the requests that arrive on any of the service’s public IP addresses. Wackamole is a completely distributed software solution based on a provably correct algorithm that negotiates the assignment of IP addresses among the available servers upon detection of faults and recoveries, and provides N-way fail-over, so that any one of a number of servers can cover for any other. Using a simple algorithm that utilizes strong group communication semantics, Wackamole demonstrates the application of group communication to address a critical availability problem at the core of the system, even in the presence of cascading network or server faults and recoveries.