Abstract: In Distributed Data Stream Management Systems，operators are executed in different processing nodes.It is a critical issue to determine operator placement in Distributed Data Stream Management Systems.An operator placement algorithm based on latency space and spring relaxation technology is proposed by Peter et al.But the algorithm assumes that the data race between two operators is constant and doesn’t consider the relativity between the data race of data streams and operator.Therefore this paper analyses the relation between the operator and its output race of data stream，and improves upon the algorithm proposed by Peter et al.Experimental results show that the algorithm can be applied to distributed data stream management systems effectively.