Parallel Database System Load Balancing Technology

Parallel database systems based on shared nothing structure is a effective solution for massive data management since it can provide sufficient computing and storage resources. Load balancing is a key issue for query processing performance in parallel database systems, we conduct the load balancing problem in this paper.The load balancing techniques in parallel database systems include static load balancing and dynamic load balancing. The decluster strategies and data reorganization of database is the effective static load balancing techniques in parallel database systems. This paper mainly studys the technologies of decluster strategies, data reorganization techniques and dynamic load balancing methods.The decluster strategies have great impacts on the query processing performance in parallel database. The existed decluster strategies selection algorithms output an optimal database decluster plan with a given static query workload, however, it isn't suitable for OLTP database in which the query workloads vary with time. This paper proposes an adaptive technique to tune the database decluster plan with the varying workloads, which can keep the optimal database decluster during the lifetime of database. A novel decluster strategy RCMD is also presented in this paper.Changing parallel database decluster can cause the expensive data reorganization. The existed methods decrease the query processing performance sharply during the data reorganization period. We propose several novel data migration and data reorganization algorithms and query processing methods in the reorganization period, which can get the optimal performance.Dynamic load balancing strategy is an effective method for the skewworkload during query processing. We propose join and aggregation algorithms based on the dynamic load balancing strategy, which can rebalance the workload among the processors during the running time of the algorithm. Theoretical analysis and experiment results show that the algorithms proposed in this paper are effective.