Object Migration in the Guide System

Object migration is a strong requirement for distributed persistent object stores. The purpose of object migration is not only to allow disk location modification for objects managed in the permanent store, but it is also often used as a means for gathering objects in some partitions also called clusters, in order to improve object management at the system level. It therefore provides the system and the applications with the ability to use their knowledge about inter-object relations for the management of clusters as some kinds of object working-sets (as in [3]). Unfortunately, existing implementations of object migration incur an important overhead on naming functions (e.g. location, addressing). In the Guide system, we have designed and implemented a mechanism for object migration between clusters that avoids these drawbacks. The advantages of our location/migration policy may be summarized as follows: our uniform naming scheme facilitates distributed concurrent sharing (as opposed to local identifiers), object location is simple and efficient for objects that don’t move, and the migration scheme reduces the location cost when correlated objects are gathered in clusters since it avoids extra mapping operations for migrated object locations. This paper presents the proposed location/migration mechanisms, their integration in the Guide object-oriented distributed system, a performance evaluation based on analysis and a short comparison with related work.