TRAFFIC MANAGEMENT IN INTEGRATED SERVICES NETWORKS: SCHEDULING AND RESOURCE PARTITIONING Traffic characterization, admission control and packet scheduling are important traffic management functions needed for provisioning QoS in Integrated Services Networks. In this thesis, we examine each of these functions in the context of generic networks and virtual (private) networks providing integrated services. We characterize burstiness of several long traces of video at different time scales using burstiness function. We examine video compressed using JPEG, MPEG and NV's compression algorithms. JPEG video has low short term burstiness and a large long term burstiness, MPEG video has burstiness at short as well as long time scale whereas NV's traffic has only short term burstiness. We discuss the implications of burstiness at different time scale on bandwidth and buffer allocation. We then focus on deterministic guarantees (which provide a bound on worst case behavior) and compare the performance of leaky bucket traffic model with $\Xmin, \Xavg I$ traffic model while using optimal admission control tests. We found that the leaky bucket model outperforms the $\Xmin, \Xavg, I$ model. We show that using good traffic models and optimal admission control tests, even deterministic guarantees provide reasonable network utilization. An important contribution of this thesis is a new scheduling algorithm called the Recursive Round Robin (RRR) scheduler. The proposed scheduler operates on fixed sized packets and needs simple bit manipulation operations to make scheduling decisions. Thus it can operate at very high speeds. We discuss several variants of the basic scheduler, including the variable size packet scheduler. The delay and fairness properties of the scheduler are analytically derived and discussed in detail. Efficient hardware and software implementations of the scheduler are also suggested. The scheduler has applications in ATM network interface cards, ATM switches and IP routers. We next show why traditional schedulers are inappropriate for providing bounded delay services in virtual (private) networks. We define the concept of output burstiness and show how generic latency-rate schedulers with bounded output burstiness can provide good end-to-end delay bound in virtual networks. We suggest two variants of the RRR schedulers for virtual networks. The output burstiness constraint reduces the amount of sharing possible in virtual networks. We propose a capacity resizing approach to improve sharing in virtual networks. The capacity increase requests which are sent at a granularity of session arrivals, are admission controlled, using a new technique called Stochastic Fair Sharing (SFS), which fairly redistributes the free link capacity among different traffic classes. SFS uses the concept of trunk reservation to give high priority to traffic classes with low normalized usage. SFS can be used to achieve better link sharing in virtual networks. We present simulation results for a single link and a twelve node network. SFS achieves fair sharing on a single link and max-min fair sharing in a network with fairness index between 0.96 and 0.99, as compared to an index between 0.66 and 0.97 for simple schemes like FCFS. The bandwidth penalty for using SFS was less than $5$\% and the signaling load for SFS in a network was less than 100 messages per second per router.