ABSTRACT

In addition to fault tolerance, replication reduces access latency. DNS servers at the upper levels are highly replicated-without this, IP address lookup will be unacceptably slow. Accordingly, in large systems (like P2P systems and grids), how many replicas will be required to bring down the latency of access to an acceptable level and where to place them are interesting questions. Another major problem in replication management is that of replica update. e problem does not exist if the data are read-only, which is true for program codes or immutable data. When a replica is updated, every other copy of that data has to be eventually updated to maintain coherence. However, due to the nite computation speed of processors and the network latencies involved in updating geographically dispersed copies, it is possible that even aer one copy has been updated, users of the other copies still access the old version. What inconsistencies are permissible and how coherence can be restored are important issues in replicated data management.