|
|
|
This paper presents serializable protocols for lazy replication of updates in a distributed database. Lazy replication means that the transaction updates a single copy, and the update is propagated to the other replicas after the completion of the original transaction. Lazy replication increases concurrecy, but until recently it was thought that it cannot guarantee serializability. A paper by Chundi, Rosenkrantz, and Ravi in icde'96 [2] has shown that if a certain graph constructed based on the replicas is acyclic, then lazy replication and serializability can be reconciled. The graph that has to be acyclic is the underlying graph of a directed one called the "copy-graph". This paper extends the Chundi, Rosenkratz, and Ravi results to the case where the the copy graph, rather than the underlying one, is acyclic. The paper describes a careful and well thought out performance analysis of the algorithms in a real system (not simulation).
The transaction model is limited in the sense that it cannot support a transaction that updates items that have primary copies at two different sites.
Overall, I found the paper to be technically interesting, combining theoretical and experimental results in an elegant framework. It makes a tangible, well articulated contribution in a "crowded" concurrency control area.
Copyright © 2000 by the author(s). Review published with permission.