A Workload Assignment Strategy for Efficient ROLAP Data Cube Computation in Distributed Systems

摘要

Data cube plays a key role in the analysis of multidimensional data. Nowadays, the explosive growth of multidimensional data has made distributed solutions important for data cube computation. Among the architectures for distributed processing, the shared-nothing architecture is known to have the best scalability. However, frequent and massive network communication among the processors can be a performance bottleneck in shared-nothing distributed processing. Therefore, suppressing the amount of data transmission among the processors can be an effective strategy for improving overall performance. In addition, dividing the workload and distributing them evenly to the processors is important. In this paper, the authors present a distributed algorithm for data cube computation that can be adopted in shared-nothing systems. The proposed algorithm gains efficiency by adopting the workload assignment strategy that reduces the total network cost and allocates the workload evenly to each processor, simultaneously.