Providing High Availability in Very Large Workkow Management Systems 1

Work ow management systems (WFMS) support the modeling, coordinated execution and monitoring of business processes within an organization. In particular, very large work ow management systems are used in organizations with several thousand users, hundreds of thousands of process instances, and several thousand sites, all distributed over wide geographic areas. In these environments, failure of the WFMS or the underlying work ow database which stores the meta-information about the processes is not tolerable. This paper addresses the problem of providing high availability in work ow management systems by proposing a backup technique which ensures that execution of a process instance can be resumed at any point in time in the event of a failure. An essential characteristic of our backup scheme is that it allows the user to de ne di erent availability levels, reducing the cost of maintaining backups. The backup scheme is implemented using the work ow semantics, which we believe will (i) make it independent of the underlying work ow database, thus permitting the use of heterogeneous databases as primary and backup, (ii) reduce overheads, especially when compared to backup schemes provided by database systems.