||With the fast development of the digital technology, both traditional and new multimedia communication services are expected to be delivered over the same network infrastructure. Based on the TCP/IP protocol suite, the Internet provides a connectionless, best effort, end-to-end packet delivery service with ubiquitous global connectivity. There is a common belief that the Internet has become the dominant networking technology and thus the need for multimedia communication with Qualify of Service (QoS) guarantee demands substantial changes in the current Internet infrastructure. Multimedia services demand not only high bandwidth, but also a stringent real-time delay constraint. We study the problem of QoS guarantee for different service classes. We employ a 2-level hierarchical scheduling framework for the separation of QoS guarantee such as bandwidth, delay and packet loss. The higher-level link sharing scheduler allocates link bandwidth among classes and the lower-level scheduler, a Generalized Processor Sharing (GPS) scheduler or an Earliest Deadline First (EDF) scheduler, is selected for the in-class scheduling service based on traffic characteristics. Due to the good property of minimizing the maximum lateness of packets, the Earliest Deadline First (EDF) scheduling is adopted to provide the in-class scheduling for the real-time traffic in this thesis, which is normally time-sensitive. Real-time applications are commonly classified as either soft or hard real-time. Soft-real-time applications can tolerate some amount of lost messages while hard-real-time applications have zero loss tolerance. Therefore, the former can be carried by the unreliable transport (UDP) while the later by the reliable transport (TCP) in the Internet. The hard-real-time applications running as TCP flows need to compete for resources with the aggressive real-time UDP flows unless they are differentiated by the networks. It is well known that the aggressive UDP sources can take advantage to get more bandwidth when competing with responsive TCP sources. In the thesis, we propose to use a EDF scheduler combined with an active buffer management scheme (CHOKe) to improve the fair resource allocation between UDP-based and TCP-based real-time applications and at the same time to maintain a good delay performance for all real-time applications. In addition, the soft-real-time applications running on the UDP transport may significantly have different loss requirements. Therefore, a novel modified EDF scheduling scheme (E-LOSS) is proposed, in which the loss probabilities of different flows are measured on-line and used for packet scheduling when the queue length exceeds some pre-specified threshold. Simulation results show that the proposed EDF scheduling working with the active buffer management scheme can achieve a better delay performance and at the same time make a more fair bandwidth allocation between real-time TCP and UDP connections than the First Come First Serve (FCFS) scheduling with the Drop-Tail buffer management commonly deployed in the traditional IP router. The simulation studies also show that the proposed E-LOSS scheme can support UDP real-time connections with different delay and loss requirements with better efficiency than the generic EDF scheduling and the modified EDF scheduling for finite buffer (EDF-FB).