As the Internet users grow, new network technologies are emerging. Those include ADSL and CATV Internet, which essentially provide asymmetric bandwidth for uplink and downlink to the user's connection. In this paper, we investigate the behavior of HTTP/TCP protocols on such asymmetric networks, and present the analytic results of the mean throughput of TCP. The transfer time of Web documents by HTTP over TCP is also derived. In the analysis, we consider newer HTTP/TCP protocols, HTTP/1.1 and TCP Vegas, in addition to HTTP/1.0 and TCP Tahoe. We then investigate the appropriate combination of HTTP and TCP protocols on the asymmetric network. The results show that the effect of HTTP/1.1 is quite small, but TCP Vegas can improve the performance in asymmetric networks if it is appropriately modified as in our proposal.

The population using Internet is increased rapidly in the recent years. Thus the network bandwidth is not enough accordingly causing the information transmitted by slowly. Therefore, the function of proxy server is more and more important. The cluster proxy server system is composed of many independent proxy servers which can enlarge the disk cache by combing each independent proxy server and balance them. Cluster proxy servers can provide a higher cache `hit rate' than any independent proxy server. In the traditional cluster proxy server schemes such as Internet cache protocol or Hashing Routing Protocols, clients can not select proxy server location dynamically and it will increase the time of searching the related documents. In this paper, a new cluster proxy server structure is proposed and called the dispatching cluster proxy server system.

We present a simulation study of WWW traffic over a DQDB network running the TCP/IP protocol stack. All network and protocol elements have been implemented in detail. A realistic DQDB network with 12 nodes and 24 access units is studied. QoS parameters of 552 TCP connections between end users are analyzed. Our traffic source model and its parameters has been chosen according to the characteristics of typical WWW traffic. We have analyzed several network models with different kinds of buffer management in the DQDB access unit. The influence of the buffer management on QoS parameters and transmission performance has been examined. In this paper we describe and compare the results of our investigations for two types of the access unit; with separate buffer management and with shared buffer management. Simulation results for the throughput of the DQDB busses and the loss of IP packets in the DQDB network are given. TCP throughput and transmission times at two protocol levels are measured. Our results show a good utilization of the DQDB network for both types of access units studied.

The mouth-to-ear delay bounds that can be tolerated for undistorted voice are well known and standardized in the ITU-T Recommendations. In this paper, similar delay bounds are determined for voice that is transported in compressed form over an IP network. More precisely, the dependency of these bounds on the low bit rate codecs used, the amount of echo control performed and the way the IP network is accessed, is investigated in detail.

Switching and scheduling are the two key elements to enable QoS awareness in IP routers. This paper extends our previous research work in ATM networks and introduces per-flow switching and per-flow scheduling in the design of QoS- capable IP routers. We switch IP packets on a per-flow basis and essentially create a virtual pipe from source to destination for every IP flow. We present a per-flow scheduling discipline to enable elasticity from every virtual pipe. We evaluate several key QoS parameters: delay, delay jitter, throughput, and packet loss ratio, to gauge the performance of our proposed switching and scheduling schemes.

In this paper, we introduce Lazy Packet Discard (LPD), an AAL-level enhancement that improves effective throughput, reduces response time, and minimizes wasted bandwidth for TCP/IP over ATM. In contrast to the SCD and EPD policies, LPD delays as much as possible the removal from the network of cells belonging to a partially communicated packet. LPD preserves network bandwidth by keeping such cells alive and by ensuring that additional cells, obtained through Reed- Solomon block coding at the sender's AAL, are eventually transmitted to salvage the packet in question. We outline the implementation of LPD and show the performance advantage of TCP/LPD, compared to plain TCP and TCP/EPD through analysis and simulations.

In this article we present an overview of the progress made in the analysis of self-similar traffic models, aggregation of several self-similar traffic streams and in particularly the queue performance using chaotic maps. We found out that the asymptotic behavior of the queue is a function only of the tail of the ON active periods and that the Hurst parameter is not a good parameter to achieve traffic control due to the fact that two different self-similar traffic traces can have the same Hurst parameter but have a very different effect on the queue statistics. These results are part of a framework for developing chaotic control of networks.

In this paper we propose an approximation for individual overflow moments of a multiservice link with differing arrival rates, capacity requirements and mean holding times, where trunk reservation is used. The approximation is a generalization of Roberts' well-known approximation for individual blocking probabilities of a multiservice link to higher moments. It can be computed very efficiently. The quality of the approximation for the second moment (variance) is comparable to Roberts' approximation for the individual blocking probabilities. Thus the results provide an efficient algorithm for computing the two moment characterization of the individual overflow streams and hence can be used for the design and analysis of circuit switched alternate routing networks with trunk reservation links.

A particular closed queuing network consisting of two processor sharing servers and two types of customers with fixed routes, which are generated by two finite sources, is studied. A complete bottleneck classification is provided as the number of customers and service rates at processor sharing servers increase. Asymptotic approximations for the marginal distributions at the two processor-sharing nodes are derived, when both nodes are bottlenecks. These approximations imply the normal approximation whose mean and variance are explicitly expressed through network parameters. The results are applied to the problem of admission control in packet-switching communication networks.

A shared buffer ATM switch with a single hot-spot destination port, which is loaded with bursty input traffic, is modeled by a discrete-time queuing system. An approximation method to analyze the queuing system under consideration is developed. We first propose an efficient aggregation algorithm for superposing all the individual input processes to the switch. Then, the shared buffer space of the switching system is partitioned into two destination parts: an address queue dedicated to the hot-spot and remaining address queues having balanced traffic loading. By using an approximation method for estimating the number of nonempty queues in the remaining address queues at an arbitrary time slot, we can obtain the steady-state probability distribution of the queuing system. From the obtained steady-state probabilities, some performance measures such as cell loss probability in the shared buffer and that in the address queue dedicated to the hot-spot can be derived. To eliminate the starvation effect of buffer hogging, we propose two traffic management policies: priority access and cell dropping strategy. Numerical examples of the proposed method are given, which are compared with simulation results.

In times of Internet access being a popular consumer applications even for `normal' residential users, some telephone exchanges are congested by customers using modem or ISDN dial-up connections to their Internet Service Providers. In order to estimate the number of additional lines and switching capacity required in an exchange or a trunk group, Internet access traffic must be characterized in terms of holding time and call interarrival time distributions. In this paper, we analyze log files tracing the usage of the central ISDN access line pool at University of Stuttgart for a period of six months. Mathematical distributions are fitted to the measured data and the fit quality is evaluated with respect to the blocking probability caused by the synthetic traffic in a multiple server loss system. We show how the synthetic traffic model scales with the number of subscribers and how the model could be applied to compute economy of scale results for Internet access trunks or access servers.

Correlated Interarrival time Process (CIPP) has been proposed, for modeling both the composite arrival process of packets in broadband networks and the individual source modeling. The CIPP--a generalization of the Poisson process- - is a stationary counting process and is parameterized by a correlation parameter `p' which represents the degree of correlation in adjacent interarrivals in addition to `(lambda) ' the intensity of the process. In this paper, we present the performance modeling of VBR video traffic in ATM networks, using CIPP/M/1 queue. We first give the expressions for stationary distributions for CIPP/M/1 queue. The, we derive the queuing measures of interest. We simulate a queue with smoothed VBR video trace data as input (with exponential services) to compare with the theoretical measures derived above. Experimental results show that the CIPP/M/1 queue, models well with ATM multiplexer performance with the real world VBR video traffic input.

This paper demonstrates that the frames of different types are highly correlated and present a novel model for MPEG video source. The model is based on characteristics of correlated traffic. Instead of using three independent stochastic processes, a single process is used and the outcomes for different frame types are obtained by means of transformations.

A new idea of design on an Integrated IP Telephone Gateway is stated in this paper. Some key techniques are analyzed and discussed in detail. A Centered Echo Chancellor based on TMS320C6201 DSP is implemented, and many optimization methods for voice compression code are concluded to enhance the performance of the gateway. Finally, the performance parameters of the Integrated IP Telephony Gateway are summarized in this paper.

In this paper, a two timescale simultaneous perturbation stochastic approximation algorithm is developed and applied to closed loop rate based available bit rate flow control. The relevant convergence results are stated and explained. Numerical experiments demonstrate fast convergence even in the presence of significant delays and a large number of parameterized policy levels.

The fair distribution of bandwidth to different connections is an important issue in high-speed networks. This is especially true in low-priority services where the bandwidth available for the low priority connections may be small and may vary rapidly. An example is the Available Bit Rate (ABR) service in Asynchronous Transfer Mode networks. This study uses the ABR service to explore a scheme that achieves better fairness in bandwidth allocations. Traditional schemes fail to achieve desired bandwidth allocations, which are generally based on connection weights. In this paper, a new scheme called Active Fairness is proposed which substantially improves fairness in bandwidth allocations. Contrary to one set of weights used in traditional schemes, Active Fairness maintains two sets of weights at each link.

Future high-speed networks will be very complex and use many sophisticated traffic control mechanisms. This paper investigates the performance of interactive TCP services over large scale ATM networks with many TCP connections by means of simulation. A simulation environment has been developed, which allows the automatic creation of large simulation scenarios on the basis of a tabular description.

Actual measurements of high-speed traffic in communications networks argue convincingly that self-similar stochastic processes should model it. These measurements have also revealed that overall buffer packet loss decreases very slowly with increasing buffer size, in sharp contrast to traditional queuing theory models where losses decrease exponentially fast with increasing buffer size. In the paper, our problem is to analytically study overflow and loss probabilities in a queue with self-similar packet traffic and get the asymptotic lower bounds to them.

The impact of the now widely acknowledged self-similar property of network traffic on cell delay in a single server queuing model is investigated. The analytic traffic model, called N-Burst, uses the superposition of N independent cell streams of ON/OFF type with Power-Tail distributed ON periods. Delay for such arrival processes is mainly caused by over-saturation periods, which occur when too many sources are in their ON-state. The duration of these over- saturation periods is shown to have a Power-Tail distribution, whose exponent (beta) is in most scenarios different from the tail exponent of the individual ON- period. Conditions on the model parameters, for which the mean and higher moments of the delay distribution become infinite, are investigated. Since these conditions depend on traffic parameters as well as on network parameters, careful network design can alleviate the performance impact of such self-similar traffic. Furthermore, in real networks, a Maximum Burst Size (MBS) leads to truncated tails. An asymptotic relationship between the delay moments and the MBS is derived and is validated by the exact numerical results of the analytic queuing model.

We study a queuing system having a mixture of special semi- Markov process (SSMP) and Poisson arrivals as the input process, where the Poisson arrival is regarded as interfering traffic. It is shown by numerical examples that the SSMP arrivals receive worse service than Poisson arrivals, i.e., the main waiting time of SSMP customers is longer than that of Poisson customers. We also propose a model of Moving Picture Experts Group (MPEG) frame arrivals as an SSMP batch arrival process. This model captures two features of the MPEG coding scheme: (1) the frequency of I, P, and B frames in a Group of Pictures, and (2) distinct size distributions for the three frames. The waiting time of each ATM cell generated from the frames is evaluated in the numerical examples. It is found that the waiting time characteristics are rather different among some real video data.

Developments in synchronous (SDH/SONET) transmission products has led to increased use of ring structures in telecommunication networks, on account of their `self- heating' properties. Often, a set of nodes is interconnected by a `family' of such rings, which all follows the same fiber route. In this situation, each node is served by a subset of the rings in the family. For a given set of nodes and a given matrix of point-to-point traffic, we describe a design methodology that determines which rings should serve each node, and how traffic should be routed on the rings. The objective is to minimize the amount of equipment required. Whereas earlier work has addressed this problem for the case of uni-directional rings, the methodology described here can be applied to bi-directional rings as well.

There are many studies dealing with handover statistics for voice wireless calls, however handovers of multimedia traffic is still in its early stage of investigation. A major difference is that neither the Call Holding Times (CHT) nor the Cell Residence Times (CRT) can be longer assumed as exponentially distributed. In this paper, we obtain analytical expressions for a variety of handoff rates for Erlang-based call holding times and cell residence times. In particular, we present closed-form analytical results for the handoff cell arrival rate in the following cases: (1) mixtures of exponential distributions for the CHT and any distribution for the CRT, (2) mixtures of k-Erlang distributions for the CHT and exponential CRT, and (3) 2- Erlang distributions for CHT and CRT. We match parameters with the birth-and-death stochastic process model and we give some insights for simulations.

Because of the large statistical complexity of the traffic in communication networks, connection admission controls (CAC) often use only crude initial estimates of the required bandwidth for variable bit-rate connections. This is one reason why on-line measurements of the existing network traffic have been proposed. The well-known effective bandwidth formulation or recent worst case bounds could be used by CAC. On-line methods of estimating the received quality-of-serve of connections from `feasible statistics' could be used to modify the bandwidth allotment of such connections. Such bandwidth modifications occur at a time- scale between those of the packet level and connection (i.e., CAC) level.

To predict the delay between a source and destination as well as to identify anomalies in a network, it is crucial to continuously monitor the network by sending probes between all sources and destinations. It is of prime importance to reduce the number of probes drastically and yet be able to reasonably predict the delays and identify anomalies. In this paper we state and solve a graph-theoretic problem to optimally select a subset of traceroute-type probes to monitor networks.

The principles of distributed object oriented programming offer great possibilities for flexible architectures in multiple fields. In telecommunications, an architecture called Telecommunication Information Networking Architecture has been developed using these very principles. It allows telecommunication services to be implemented using software objects that in turn can be executed in a location transparent way in a network. The location transparency offers great flexibility for service creation, but as the software must be executed somewhere in the network on nodes of finite capacity, performance problems can arise due to inefficient placement of objects causing either overloaded nodes or excessive and unnecessary inter-node communication. To ensure good performance, various measures of load control and load balancing must be taken. We discuss how to measure the performance of a distributed object oriented system and examine two load balancing algorithms that can be used in such systems.