Availability Management Abstract

This paper describes the design and implementation of an automatic availability management service called Teams for a timed asynchronous distributed system. Teams automatically reconfigures a distributed system in the presence of communication and node failures in such a way that all computing services remain available, and the system reconfiguration is transparent to the users. Teams provides automatic availability properties to the computing services that use either the primary-backup model or the active replication model for replication. It is a fail-aware service in the sense that a node at any point in time knows whether it can provide the automatic reconfiguration services or not. Teams provides excellent performance and can reconfigure the system in response to node failures, communication partitions, or maintenance operations in less than 8 milliseconds.