Advertisement
Published in:
2017 | OriginalPaper | Chapter
Replica-Aware Partitioning Design in Parallel Database Systems
Authors : Liming Dong, Weidong Liu, Renchuan Li, Tiejun Zhang, Weiguo Zhao
Published in: Euro-Par 2017: Parallel Processing
Publisher: Springer International Publishing
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
Abstract
In parallel database systems, data is partitioned and replicated across multiple independent nodes to improve system performance and increase robustness. In current practice of database partitioning design, all replicas are uniformly partitioned, however, different statements may prefer contradictory partitioning plans, so a single plan cannot achieve the overall optimal performance for the workload.
In this paper, we propose a novel approach of replica-aware data partitioning design to address the contradictions. According to the access graph of SQL statements, we use the k-medoids algorithm to classify workload into statement clusters, then we use the branch-and-bound algorithm to search for the optimal partitioning plan for each cluster. Finally, we organize replicas with these plans, and route statements to their preferred replicas. We use TPC-E, TPC-H and National College and University Enrollment System (NACUES) to evaluate our approach. The evaluation results demonstrate that our approach improves system performance by up to 4x over the current practice of partitioning design.