Access skew is the most important challenge for scalable parallel systems, especially when data are in range partitioned schema. To realize scalability, many dynamic access skew balancing methods with data reorganization and parallel index structures on shared-nothing parallel infrastructure have been proposed. Data migration with range-partitioned placement using a parallel Btree is one solution. The combination of range partitioning and chained declustered replicas provides the dynamic skew balancing abilities for high scalability. However, no previous study has provided any practical implementation of this capability. In addition, independent treatment of the primary and backup data in each node results inefficient skew balancing. We propose a novel compound parallel index, termed Fat-Btree, to provide access paths to both primary and backup data across a chained declustering system for efficient dynamic skew balancing with low maintain cost. Experiments using PostgreSQL on a 160-node PC cluster demonstrate the effects.
各種検索
サポート
ヘルプ