Recent measurements on LAN, MAN and WAN traffic have demonstrated that Long-Range Dependence (LRD) is an invariant property irrespective of network technology being employed. As a consequence, the performance of the network is dominated by the property of LRD in the network traffic. Latency in information access is one of the most important factors in the user perceived Quality of Service (QoS) in network applications. Almost all the applications follow the client-server paradigm to transfer information entities which are typically files or typed-in messages. The distribution of the sizes of these information entities are best described by heavy-tailed distributions and results in LRD. This fundamental property impacts all the aspects of application layer performance (e.g., response time) and the network layer performance (e.g., packet loss and delay). In the traditional traffic models, the network resource management is aimed at capturing the timescales at which bursts occur and dimensioning the network for the burst sizes occuring at that timescale. This management paradigm has proven to be ineffective for the traffic having significant level of LRD. Because of LRD, bursts of all possible sizes occur at timescales spanning over several orders of magnitude. Engineering of network resources to protect application layer QoS is therefore an important task. In this paper heavy-tailed distributions are used to model the information contents transferred by some of the classical network applications such as FTP, SMTP and HTTP. The parameters of these models are based on high resolution non-intrusive monitoring of busy periods in live networks. The clients and the servers are modelled as ON-OFF sources producing LRD phenomena at the packet level through aggregation. The user level quality for these applications is investigated (in terms of end-to-end delay performance) and preliminary results are reported showing how the quality is affected by the bandwidth and buffer allocation schemes.