A Hierarchical Replica Location Approach Based On Cache Mechanism And Load Balancing

A data grid connects a collection of geographically distributed computers and storage resources that may be located in different parts of a country or even in different countries, which enables users to share data and other resources and collaborate with each other.In data grid, replication of remote data can avoid access bottleneck, and help to reduce access latency and bandwidth consumption.To do this, replica location service is required in data grid environments. A Replica Location Service is a core component of one grid data management architecture, which provides a mechanism for registering the existence of replicas and discovering them according to given parameters.In data grid environments, data is heterogeneous, distributes widely around internet, data-sharing characteristics exploit small-world characteristics, which are temporal locality, geographical locality and spatial locality. Considering these characteristics and p2p objection location technology, we present a novel replica location approach, a hierarchical replica location approach based on Cache mechanism and load balancing, named RepliLoc.In RepliLoc, we focus on two issues: storage load balancing and computational load balancing. Storage load balancing is achieved via rough evenly distributing replica index maps among nodes by using DHT and prefix matching.Computational load balancing can be accomplished via localizing replica location by data community. Replica location is classified into two layers: replica location on community layer and community alliance layer. On community layer, we present a circular information dissemination mechanism based on small world model (WS model) to accomplish replica location. On the other layer, we propose a replica location approach based on prefix matching and DHT. In addition, Cache mechanism is used by both the two layers to shorten query latency. A series of experiments proved that our approach is efficiency and effectiveness.