chain replication-osdi.pdfVIP

Chain Replication for Supporting High Throughput and Availability Robbert van Renesse Fred B. Schneider rvr@ fbs@ FAST Search Transfer ASA Troms?, Norway and Department of Computer Science Cornell University Ithaca, New York 14853 Abstract pre-programmed, possibly non-deterministic, computation involving the prior state of that Chain replication is a new approach to coordinating object. clusters of fail-stop storage servers. The approach is intended for supporting large-scale storage services A ?le system write is thus a special case of our stor- that exhibit high throughput and availability with- age service update which, in turn, is a special case out sacri?cing strong consistency guarantees. Be- of a database transaction. sides outlining the chain replication protocols them- Increasingly, we see on-line vendors like Ama- selves, simulation experiments explore the perfor- , search engines like Google’s and mance characteristics of a prototype implementa- FAST’s , and a host of other information-intensive tion. Throughput, availability, and several object- services provide value by connecting large-scale stor- placement strategies including schemes based on age systems to networks. A storage service is the distributed hash table routing are discussed. appropriate compromise for such applications, when a database system would be too expensive and a ?le system lacks rich enough semantics. 1 Introduction One challenge when building a large-scale stor- age service is maintaining high availability and A storage system typically implements operations high throughput despite failures and concomitant so that clients can store, retrieve, and/or change changes to the storage service’s con?guration, as data. File systems and database systems are per- faulty components are detected and replaced. haps the best known examples. With a ?le s