||Differentiated services (Diffserv) is a new service model proposed by the Internet Engineering Task Force (IETF) for the Internet to enable the provision of an end-to-end quality of service (QoS) support for multimedia applications. A small number of per hop behaviors (PHBs) are defined to characterize the forwarding behavior. Packets are classified, marked and policed on entry and assigned to a particular Behavior Aggregate (BA) depending upon the Service Level Agreement (SLA). Flow aggregates are provided with forwarding behavior based on the BA they belong to. Since nodes inside the network need not maintain per flow information, it scales very efficiently. The key elements of the Diffserv for supporting QoS include the definitions of different PHBs like Expedited Forwarding (EF) and Assured Forwarding (AF). EF can be used to build a low loss and low delay service to support applications like Internet telephony or video conferencing. In this dissertation, first the performance characteristics of the EF PHB implementation using Class Based Queuing (CBQ) are studied. It is observed that EF performs better when implemented using CBQ in comparison to when weighted round robin (WRR) is used. The effect of packet size, number of flows, and service ratio is also studied. Further, the performance of Voice over Internet Protocol (VoIP) over Diffserv using EF is studied, in the presence of Best Effort (BE) traffic. Results are compared for the delay and jitter performance of the traffic generated by different standard voice codec algorithms used for compressing a voice signal. Best effort traffic sources are exponential on-off or long term TCP. All experiments are carried out for two different topologies i.e. dumbbell and linear, and for undersubscribed and oversubscribed network conditions. It is observed that in homogeneous flow aggregates consisting of VoIP traffic from a given voice codec, a small packet sized flow experiences shorter delay and jitter compared to flow aggregates consisting of large packet sized flows. It has also been observed that in a homogeneous flow aggregate, the jitter experienced by an individual flow decreases as packet size decreases, as the number of flows increases in a homogeneous aggregate, the jitter experienced by an individual flow increases. It is also observed that a change in packet size affects the jitter much more than a change in the number of flows in a homogeneous flow aggregate. In heterogeneous flow aggregates, consisting of VoIP traffic generated by different voice codecs, jitter experienced by small packet sized flows are longer in comparison to that experienced by large packet sized flows; and as the number of flows increases in a heterogeneous flow aggregate, the jitter and the delay experienced by an individual flow also increases. For an undersubscribed network, trends for delay and jitter are the same as that of an oversubscribed network, except that the absolute values are smaller. Results are similar for both the dumbbell topology and linear topology. It is also observed that the delay and jitter experienced by VoIP flows are longer when Best Effort traffic is generated by long term TCP sources, than that experienced in the presence of BE traffic generated by exponential on-off sources.