Efficient Parallel Multi-way Merging on Heterogeneous Multi-core Cluster

Abstract

Data merging and sorting are often applied to the scientific and engineering applications such as computational fluid dynamics and computation geometry. By constructing a data sending matrix, solving the data exchange range and determining the data exchange order among compute nodes to reduce the communication overhead, this paper proposes a load-balance data distribution strategy among nodes, and designs a communication-efficient parallel multi-way merging algorithm on the heterogeneous cluster with the multi-core compute nodes which have different computation speed, communication rate and memory capacity. The experimental results on the heterogeneous cluster with multi-core machines show that the proposed parallel merging algorithm obtains high speedup and has good scalability.