De ning and Measuring Con icts in Optimistic Replication

Optimistic replication is often viewed as essential for large scale systems and for supporting mobile computing. In optimistic replication, updates can be made concurrently to di erent le replicas, resulting in multiple versions of the le. To recover from these con icting updates, after-the fact con ict resolution actions are required to recombine multiple versions into one. This paper de nes these concepts and discusses approaches to measure them in optimistically replicated systems. Measurement of the number of con icting updates and con ict resolution is important to judge the practicality of optimistic replication. An environment where con icting updates are frequent will not be attractive since users cannot assume they have up-to-date data. Although many con icts can be automatically resolved, some con icts require user intervention; such con icts cannot be too common. This paper shows an approach to measure the number of con icting updates. From this measurement we derive the actual amount of work done by the user or system to resolve con icts and the minimum amount of work required to resolve con icts.