higher level performance, reliability and availability. However,
advantages of replication do not come without a price. Scalability of
existing replication database techniques that guarantee database
consistency presents a major challenge in the current distributed
environment with thousands of data nodes geographically distributed
among nodes of communication network.
In this talk we outline methods and algorithms and discuss results of
performance study of these methods. Our methods are based on lazy
replica update protocols which propagate updates through independent
transactions after the original transaction commits. If lazy protocols,
however, are used indiscriminately, they may result in a loss of data
consistency. We discuss four lazy replication propagation methods that
guarantee data consistency and are scalable in the database environment
with 100's data nodes.
First two protocols do not impose any restriction on placement of data
replicas but require availability of broadcast to maintain a consistent
view of replica processing by each data node. Our next two methods
require some restriction on replica placement but do not require network
broadcasts. Finally we consider hybrid protocols that do not restrict
replica placement and do not require n etwork broadcasts as well.
Results of this work were obtained in cooperation with Rastogi,
Seshadri, Korth, and Silberschatz.