Towards self-tuning data placement in parallel database systems

Abstract

Parallel database systems are increasingly being deployed to support the performance demands of end-users. While declustering data across multiple nodes facilitates parallelism, initial data placement may not be optimal due to skewed workloads and changing access patterns. To prevent performance degradation, the placement of data must be reorganized, and this must be done on-line to minimize disruption to the system. In this paper, we consider a dynamic self-tuning approach to reorganization in a shared nothing system. We introduce a new index-based method that faciliates fast and efficient migration of data. Our solution incorporates a globally height-balanced structure and load tracking at different levels of granularity. We conducted an extensive performance study, and implemented the methods on the Fujitsu AP3000 machine. Both the simulation and empirical results demonstratic that our proposed method is indeed scalable and effective in correcting any deterioration in system throughput.