Understanding the variability of Internet traffic in backbone networks is essential to better plan and manage existing networks, as well as to design next generation networks. However, most traffic analyses that might be used to approach this problem are based on detailed packet or flow level measurements, which are usually not available throughout a large network. As a result there is a poor understanding of backbone traffic variability, and its impact on network operations (e.g. on capacity planning or traffic engineering).This paper introduces a metric for measuring backbone traffic variability that is grounded on simple but powerful traffic theory. What sets this metric apart, however, is that we present a method for making practical measurements of the metric using widely available SNMP traffic measurements. Furthermore, we use a novel method to overcome the major limitation of SNMP measurements -- that they only provide link statistics. The method, based on a "gravity model", derives an approximate traffic matrix from the SNMP data. In addition to simulations, we use more than 1 year's worth of SNMP data from an operational IP network of about 1000 nodes to test our methods. We also delve into the degree and sources of variability in real backbone traffic, providing insight into the true nature of traffic variability.

We describe a preferential admission control and load balancing algorithm for distributed Internet server clusters. Clients initiate sessions consisting of a series of transactions. The clients may be aware of the various clusters of the distributed server and have preferences as to which cluster should process their session request. The scheme consists of a dispatcher which receives session requests and either admits or rejects those requests. Admitted requests are routed to their preferred cluster when the cluster is not congested. The algorithm also handles the case where a number of clients served have no cluster preference. We describe simulation results which demonstrate that the algorithm provides effective session admission control and load balancing, while maximizing the number of clients preferred by their most preferred cluster.

In this paper we focus on real-time scheduling of "soft" real-time data services such as multimedia data, MPEG video streaming and IP telephony, which can tolerate a small degree of loss or delay. We argue that network operators and service providers should be able to select from a range of quality of service objectives, including maximizing the number of customers receiving good service. Further, we argue that scheduling disciplines such as fair queueing are unable to achieve such goals and hence there is a need for alternative approaches. We propose a new scheduling scheme, which we call the dual queue discipline. We show that the dual queue has the flexibility to satisfy a variety of QoS objectives, ranging from existing notions of fairness through to maximizing the number of customers receiving good service. In addition, even the simplest dual queue implementation outperforms fair queueing, is scalable in the number of active sessions, and can be made fair, if desired, over moderate to long time scales.

The interest sparked by observations of long-range dependent traffic in real networks has lead to a revival of interest in non-standard queueing systems. One such queueing system is the M/G/1 queue where the service-time distribution has infinite variance. The known results for such systems are asymptotic in nature, typically providing the asymptotic form for the tail of the workload distribution, simulation being required to learn about the rest of the distribution. Simulation however performs very poorly for such systems due to the large impact of rare events. We provide a method for numerically evaluating the entire distribution for the number of customers in the M/G/1 queue with power-law tail service-time. The method is computationally efficient and shown to be accurate through careful simulations. It can be directly extended to other queueing systems and more generally to many problems where the inversion of probability generating functions complicated by power-laws is at issue. Through the use of examples we study the limitations of simulation and show that information on the tail of the queue-length distribution is not always sufficient to answer significant performance questions. We also derive the asymptotic form of the number of customers in the system in the case of a service-time distribution with a regularly varying tail (e.g. infinite variance) and thus illustrate the techniques required to apply the method in other contexts.

The ITU-T (International Telecommunications Union, Telecommunications Standardization Sector, formerly known as CCITT), in its specification of the SS7 protocol, describes a number of alternatives for congestion control. In this paper we provide a direct comparison of the effectiveness of some of these methods in dealing with a mass call-in situation. We find that the international option provides better call throughput and is more robust to threshold settings and delays in feeding back congestion information than the national option with congestion priorities and a combination of the two options. The results can provide guidance to those attempting to define application level overload procedures, to those studying internetworking issues for connection between countries employing different SS7 congestion control options, and to those preparing to formulate B-ISDN overload strategies.>

The author considers the performance of a Signaling System Number 7 network when the routing processors, as opposed to transmission facilities, of signaling transfer points are overloaded. The choice of overload controls in such a situation is implementation-dependent, with an option of simply discarding messages in excess of the signaling transfer point (STP) processing capability. It is this option that is studied. Call completion performance, rather than message throughput is considered as being the primary performance measure of interest since it most accurately reflects the service provided to customers. To determine realistic call completion estimates. the authors explicitly incorporate into their analysis the effects of application-level recovery procedures and customer reattempts, both of which significantly impact the service levels achieved. In so doing, they demonstrate that message throughput can be a very misleading measure of the network's ability to provide service. The need for some form of feedback mechanism to the traffic sources that will allow them to appropriately control traffic entering the network is demonstrated.>