Finding Data in the Neighborhood

摘要

In this paper, we present and evaluate alternative techniques to effect the use of location-independent identifiers in distributed database systems. Location-independent identifiers are important to take full advantage of migration and replication as they allow accessing objects without visiting the servers that created the objects. We will show how a distributed index structure can be used for this purpose, we will present a simple, yet effective replication strategy for the nodes of the index, and we will present alternative strategies to traverse the index in order to dereference identifiers (i.e., find a copy of an object given its identifier). Furthermore, we will discuss the results of performance experiments that show some tradeoffs of the proposed replication and traversal strategies and compare our techniques to an approach that uses location-dependent identifiers like many systems today.