Quality of Service(QoS) is used to define different forms of network traffic to provide different levels of service assurance capabilities. The models and architectures of QoS fall into several types. A new type of application model of QoS—Differentiated service(Diffserv) application model is proposed according to the standard of RFC2474 and RFC2475. This model is composed of sub-modules such as classifier, meter, marker(remarker), dropper and scheduler etc. Meanwhile, the algorithm of Weighted Round Robin with Max Bounded Delay(WRR-MBD) of scheduler is proposed and analyzed in detail. Finally, the implementation of Diffserv based on 3-layer switch is described. Through a network analyzer to test QoS of 3-layer switch, it is concluded that in WRR scheduling policy, the loss rate of packets with different priorities is basically allocated according to the ratio of weight, which avoids low priority queues “hunger” existing in strict priority scheduling policy.