A data transmission algorithm for distributed computing system based on maximum flow

Data skew can lead to load imbalance and longer computation time in the distributed computing system. To avoid data skew and reduce the data computation time, it is necessary to transmit the data to appropriate machines, this may however take too much network resources. How to balance the computational resources and the network resources is a problem. In this paper, we introduce a computation model called distributed two-phase model, in which the process of a task can be divided into two independent phases: data transmission and data computation. Suppose an upper bound of relative computation time is given, we show how to schedule data transmission with minimum resources, such as data transmission time and occupied bandwidth, to meet the demand. In this paper, we present a novel algorithm to minimize data transmission time and network bandwidth usage in the data transmission phase, with the conditions that an upper bound of relative computation time of data computation phase is given. Moreover, the number of nodes that participate in data computation phase is also reduced, in this way, the computational resources are saved. The simulation results show that the occupied bandwidth can be reduced effectively (about 70 %) in the situation of large-scale data sets and large number of nodes. Our algorithm is also shown to be available in replication situation.