1. Managing Update Conflicts in Bayou, a Weakly Connected Replicated Storage System D. B. Terry, M. M. Theimer, K. Petersen, A. J. Demers, M. J. Spreitzer and C. H. Hauser http://www.parc.xerox.com/csl/projec ts/bayou

2. Bayou Storage System Overview Mobile computing environment Infrastructure for collaborative applications Relying only on the weak connectivity (occasional, pair-wise communication) Supporting disconnected workgroups Weakly consistent replication Supporting automatic application -specific conflict detection and resolution at the granularity of individual update operations

3. Bayou’s Basic System Model

4. Bayou Applications Non-real-time collaborative application Examples –Meeting room scheduler Reserve meeting room for give period of time –Bibliographic database Cooperatively managing bibliographic entries Each entry has a unique, human-sensible key, such as “Jones95” or “Jones95a”

5. Conflict Detection and Resolution Observation: different applications have different notions of what it means for two updates to conflict

6. Application-specific conflicts Meeting Room Scheduler –Two meetings scheduled for the same room overlapping in time Bibliographic Database –Two entries describing different publications have been assigned the same key by their submitters –Two entries describing the same publication have been assigned distinct keys

7. System support for automatic conflict detection and resolution storage system must provide means for an application to specify its notion of a conflict along with its policy for resolving conflicts

8. Bayou’s Mechanisms for Automatic Conflict Detection and Resolution dependency check –Each WRITE operation includes a dependency check: query, expected-result –It is a precondition for performing the update that is included in the WRITE operation Merge Procedure –Each WRITE operation includes a merge proc –Written in high level, interpreted language –Producing a revised update to apply

11. Replica Consistency Anti-entropy sessions –Exchanging WRITEs –Agree on WRITEs and their order Bayou features supporting consistency –WRITEs are performed in the same, well- defined order at all servers –The conflict detection and merge procedures are deterministic

12. WRITE states Tentative –May be executed several times at the server –May produce different results Committed –has been executed for the last time at the server –Preceding WRITE log is fixed at the server

13. Global Orders of WRITEs Tentative WRITEs –timestamp is assigned to a tentative WRITE by its accepting servers –ordered according to Committed WRITEs –Committed by the primary server –ordered according to the times at which they commit and before any tentative WRITEs

14. WRITE Commit in Bayou Primary commit scheme –one designated primary server for a replicated data collection –as non-primary servers contact the primary, their tentative writes are converted to committed writes –knowledge of committed writes and their ordering propagates from the primary back to the non-primary servers via anti-entropy

15. Conclusion Weak connectivity Pair-wise client-server and server-server communication Weakly consistent replication Automatic conflict detection and resolution Global order for all update operations Moving towards eventual consistency