Abstract: At present, the load balancing method of the data center network (DCN) has some problems, such as the lack of global real-time detection for the scheduling of large and small streams. Some large flows may cause congestion, resulting in load imbalance and bandwidth fragmentation. To solve the above problems, a traffic load balancing algorithm called dynamic scheduling algorithm-delay (DSA-D) for software-defined networks (SDN) network is proposed. Firstly, the traffic is classified and the shortest hop path set of the reachable path from the source to the destination host is calculated for the flow. Then, the link delay is measured according to LLDP and ECHO to obtain the delay optimal path set. Finally, the probability fitting algorithm is used to allocate paths to realize the load balancing of data center network traffic. Experimental results in the same scenario show that DSA-D algorithm has better throughput, link bandwidth utilization and average round-trip delay than ECMP, Hedera and DIFFER algorithms.