Minimal sufficient buffer space for data redistribution in a parallel database system

摘要

A method, system and program for distributing data to a first node in a parallel database system from a plurality of existing nodes. First, a minimal sufficient number of communication buffers necessary to guarantee that a table scan operation in each of the existing nodes will execute without stopping is determined. Once calculated, the minimal sufficient number of communication buffer is allocated in the memory associated with each of the existing nodes. Thus, data from each of the plurality of existing nodes to the first node can be redistributed without interruption. Unless the redistribution process has a default set of data to be redistributed, which data from which data tables in each of the plurality of existing nodes is to be distributed to the first node must be determined. The first node may either be a new node which must be registered within the parallel database system, or an existing node within an imbalanced parallel database system which has excess capacity. The minimal sufficient number of communication buffers is determined according to the ratio of a mean queuing in a communication buffer and an average time to fill a communication buffer.