Multimedia applications require communications services with guaranteed quality of service. Connection Admission Control assures these guarantees, or makes the decision to refuse a new connection. The CAC problem is formulated as a constrained optimization problem with objective functions dependent on the class of service required. Solutions based on gathering standard ATM Performance Parameters are proposed. A Dynamic Routing Procedure is outlined, which utilizes these solutions.
A rapid growth of the Internet and proliferation of new multimedia applications lead to demands of high speed and broadband network technologies. Routers are also necessary to follow up the growth of link bandwidths. From this reason, there have been many researches on high speed routers having switching capabilities. To have an expected effect, however, a control parameters set based on traffic characteristics are necessary. In this paper, we analyze the network traffic using the network traffic monitor and investigate the Internet traffic characteristics through a statistical analysis. We next show the application of our analytical results to parameter settings of high speed switching routers. Simulation results show that our approach makes highly utilized VC space and high performance in packet processing delay. We also show the effect of flow aggregation on MPLS. From our results, the flow aggregation has a great impact on the performance of MPLS.
In this paper we will present a survey of ways to do quality differentiation and packet classification. A group of classification mechanisms are investigated more closely, namely measurement based classifiers.
Congestion control avoidance in computer networks is still a major unresolved image. The applicability of previous congestion control mechanisms has to be demonstrated taking into account today's constraints. In this work, several schemes are studied in order to support differentiated services in a wide area, very high speed network.
With the increasing focus on traffic prioritization to support voice-data integration in corporate intranets, practical methods are needed to dimension and manage cost efficient service partitions. This is particularly important for the provisioning of real time, delay sensitive services such as telephony and voice/video conferencing applications. Typically these can be provided over RTP/UDP/IP or ATM DBR/SBR bearers but, irrespective of the specific networking technology, the switches or routers need to implement some form of virtual buffer management with queue scheduling mechanisms to provide partitioning. The key requirement is for operators of such networks to be able to dimension the partitions and virtual buffer sizes for efficient resource utilization, instead of simply over-dimensioning. This paper draws on recent work at Queen Mary, University of London, supported by the UK Engineering and Physical Sciences Research Council, to investigate approximate analytical methods for assessing end to end delay variation bounds in cell based and packet based networks.
This paper presents an evaluation of various GFR (Guaranteed Frame Rate) implementation proposals. By means of extensive simulations performed in different network environments we compare two ATM Forum example implementations, namely the `simple FIFO-based GFR.2 implementation' and the `per-VC threshold and scheduling implementation'. The lessons learned from this study are as well applicable to non-ATM network technologies.
In this work, we analyze the performance of multi-class Internet traffic classifier primarily in a connection- oriented IP router environment. We define the tasks and related concepts of traffic classification in the Internet and then proceed to construct a multi-class traffic classifier using the Learning Vector Quantization algorithm classifier that has been previously used to divide the traffic into two classes. We show how the functionality of the 2-class LVQ classifier can easily be extended to an arbitrary amount of classes, in this work to three: the hard-interactive, the elastic and the best effort service classes.
In recent years, Internet traffic has been increased rapidly as a result of the Internet which accommodates multimedia traffic such as IP telephony and video conference. Gigabit routing technology is one possible approach to handle such internet traffic. This paper presents an efficient IP forwarding architecture adequate for Gigabit Ethernet switching system. The presented IP forwarding architecture is based upon distributed and pipelined process, which can effectively facilitate searching, editing, traffic classification, forwarding, and traffic management in parallel. Additionally, it can also process packets at full wire-speed in the ASIC level.
In the LCT-UC we are currently working on a new service model for Internet based communications. The main goal of this model is to provide QoS without too much complexity. For this, we need to reinvent and develop new mechanisms for routers, which effectively can differentiate the way packets are treated. This paper presents our scheduling approach, the first version of the prototype under construction, and some tests that prove its ability to efficiently differentiate traffic classes.
The Differentiated Services (DiffServ or DS) framework takes an edge over IntServ because it is scalable and lesser complex. On the other hand, the application level end-to-end quality of service, in DiffServ, may get compromised because: (1) network resources are not allocated at microflow level (a data stream pertaining to a single connection) but at aggregate level (collection of one or more microflows), (2) the DiffServ working group does not specify algorithms for PHBs but their output behaviors and (3) end-to-end quality is function of Service Level Agreements (SLAs) between the adjacent domains transited by the connection and a large diversity in SLAs is quite evident as each DS domain would have different service provision policies. We focus, in this paper, on the first two issues. Our goal is to have DiffServ deployed with all its simplicity and still be able to provide application level end-to-end quality of service. For that, we study a PHB for AF classes. A PHB comprises a packet scheduler and a packet accept/discard algorithm.
The Internet consists of a network of networks. Internet users and service provides want to provide and receive multiple services. The legacy networks till now have provided narrow bandwidth that has restricted the range of services. Asynchronous Transfer Mode (ATM) can simultaneously deliver multiple services over one network and today ATM has become a component of the Internet. An ATM switch can deliver current Internet data using UBR or ABR services. Unspecified Bit Rate (UBR) using AAL5 is the most common offering these days for data transport. Because UBR does not guarantee any QoS categories and it is a `best effort' service, cell-discarding protocols must coexist. Congestion control is always a host topic for data networks. In data networks many flow mechanisms to resolve network congestion have been proposed. Cell loss is one of the most important and critical categories for traffic management of data networking. We compare the well-known Early Packet Discard for UBR with Quantum Flow Control for ABR services with TCP over ATM. Simulation results are provided that allow a comparison of both techniques.
To overcome the inherent limitation of the single best- effort service currently provided the Internet, providing differentiated services for different classes of applications is being discussed in IETF. In this paper, we propose a lightweight packet scheduling algorithm that allocates forwarding resources to different classes, which we name controlled priority (CP) gateway algorithm. The proposed CP algorithm consists of two mechanisms denoted by CP-CQD (controllable queuing delay) and CP-STI (service time interval), respectively. CP-CQD controls the queuing delay for a class of delay/jitter-sensitive traffic. CP-STI is to service classes that require bandwidth assurance. The proposed algorithm can provide guaranteed bounds of delay, jitter, rate, and packet loss to certain aggregate flows. The CP-CQD can accommodate variable bit rate as well as constant bit rate flow without bandwidth reservation for end-to-end delay bound and minimum delay jitter. The CP-STI enables tagged classes to get guaranteed throughput. These two modules can control resources allocated to the traffic classes by adjusting parameters in response to local congestion level. The simulation results show that CP gateway algorithm can provide required quality of service to certain classes while easing the negative effects on best- effort classes.
To design, develop and validate resource allocation algorithms and configuration data for corporate networks, realistic scenarios need to be modeled and experiments carried out. This requires the use of accelerated simulation modeling tools. This paper describes a selective, multi- level approach to structuring simulation models of telecommunications networks. The multiple level range from cell/packet, burst, and call/transaction levels to recurrent trends above the call level. The appropriate range of levels being modeled is selected according to the conditions of the simulation, or the requirements of the scenario. The selective approach ensures that modeling accuracy is maintained where, and when, it is needed. Experimental results show that selective multi-level modeling gives sufficient speed increase so that large networks can be modeled within reasonable computation times. This approach forms the basis of a structure for accelerated modeling tools for evaluating corporate telecommunication network scenarios. Experimentation with such tools enables planners to configure corporate networks for service provision scenarios in order to make best use of the expensive transmission resources.
Corporate telecommunications network designers face a bewildering array of system design options. This range of options is set to increase as new terrestrial and satellite based mobile networking technologies become viable and cost effective. For example, heavily interconnected Low Earth Orbit networks provide the global capability to extend corporate VPNs to a mobile workforce. But how do network designers assess the impact of these system options given various corporate user demand scenarios? This paper draws on recent work at Queen Mary, University of London, supported by the UK Engineering and Physical Sciences Research Council, to investigate network modeling for resource management in corporate telecoms networks and satellite inter-networks supporting heterogeneous services. The paper focuses on a novel network modeling process which provides a generic capability for efficiently partitioning user demand onto network topologies.
Quality of Service (QoS) is a difficult term to define for multimedia applications. The main reason is that both audio and video quality are subjective and difficult to quantify. Much work has been done in the past to map the subjective quality of video and audio into measurable quantities. Unfortunately, when it comes to IP environments, not much experience and mathematical work exists that can be used to define robust metrics for measurement of QoS. In this paper, we report on measurements of multimedia QoS and try to map subjective criteria to discrete measurables in terms of packet loss rates, packet delays, and other quantities. We report the results of measurements done at the application level and show how network characteristics affect the perceived quality of multimedia applications. In particular, we analyze the application traffic generated by MBone clients in a distributed network education scenario.
This paper integrates a functional transport and control layer network architecture for MPLS emphasizing Traffic Engineering concepts such as the specification and provisioning of end-to-end QoS service layer agreements. MPLS transport networks are provisioned considering administrator-defined policies on bandwidth allocation, security, and accounting techniques. The MPLS architecture consists of the transport and control layer networks. The transport layer network is concerned with configuration, packet forwarding, signaling, adaptation to higher layers, and support of higher layers. The control layer network is concerned with policy configuration, management, distribution, definitions, schemas, elements, settings, and enforcement.
This paper describes and analyses a solution to the problem of data synchronization and replication for distributed entities such as directories in IP communication networks. We discuss the role of directories in the developing IP communications service infrastructure. The data replication solution we have implemented is based on the protocol specifications for the Internet, titled `Server Cache Synchronization Protocol' (SCSP). We review the requirements of using and maintaining data that is shared among many applications while the data resides in different physical locations. We give a brief description of the SCSP and discuss its implementation. We point out some possible applications for the protocol in a mixed IP/ISDN network. We also review some alternative approaches to directory services. In conclusion we propose the SCSP as a component for directory enabled networks--a concept emphasizing the key role of directories in the merging communications infrastructure. New emerging services manage large amounts of data. To facilitate the data management it is distributed over different locations following directory structures where the information is close to the customer location. The main goal is to achieve a global service accessible from everywhere, independently of the location where the user is accessing the service.
Robust-WDM is a technique to realize Wavelength Division Multiplexing (WDM) Local Area Networks (LAN's) in the presence of laser wavelength drifts. Various Medium Access Control protocols have been proposed for Robust-WDM LAN's. Among these protocols, the one with Aperiodic Reservation and Lenient Token-Passing control channel (the AR/LTP protocol) is the most promising. We discuss three internetworking strategies for AR/LTP Robust-WDM LAN's. The aim is to explore the possibility to scale the AR/LTP Robust-WDM concepts to the metropolitan domain by looking at some basic medium-access arrangements and specifying the advantages and limitations of each. Special Remote Access Nodes (RAN's) are proposed to facilitate interconnection. It is shown that by some modifications in the basic AR/LTP local area protocol, the waiting time performance of a Robust-WDM interconnection can be improved. The improvement would be accomplished at the expense of some control sophistication. Further improvement can be achieved by designing a set of point-to-point links among the RAN's of different Robust-WDM stars. In this case, control is relatively simplified, but the design of a RAN is made more complex and more expensive.
IP multicast routing between domains can be based on roles which border routers possess with regards to datagram forwarding to a particular group. We show that distributed role based routing is sufficient for the Internet backbone where topology is relatively poor. As a solution for necessary mediation between the backbone and domains the multicast mediator is proposed with multi-dimensional caching of IP multicast addresses enriched by group membership information. This brings an opportunity to co- ordinate multicast address allocation management between domain and to achieve more than best effort address availability.
The Internet has grown rapidly in the last several years. This is largely due to the simple, flexible, and robust connectionless nature of the Internet Protocol (IP). The Internet architecture has been successful up to this point, and the best effort service paradigm has been adequate. However, with increasing demands of supporting voice, video, mission critic data on IP, the best effort paradigm without differentiating traffic according to application requirements can not meet the market demands. In this paper, we summarize and compare two major models in IP Quality of Service (QoS): Integrated Services Model and Differentiated Services Model proposed in Internet Engineering Task Force (IETF), the main Internet standard committee. We also discuss other related areas in IETF, e.g., Multiprotocol Label Switching and Resource Reservation Protocol in terms of their impacts on supporting QoS in IP.
Recent advances in packet switching, internetworking, and digital signal processing technologies have converged to allow realizable practical implementations of packet telephony systems. This paper provides a tutorial on transmission engineering for packet telephony covering the topics of speech coding/decoding, speech packetization, packet data network transport, and impairments which may negatively impact end-to-end system quality. Particular emphasis is placed upon Voice over Internet Protocol given the current popularity and ubiquity of IP transport.
In IP telephony, a call is usually established in multiple stages. In the first stage, an ingress or call-originating IP-PSTN gateway (GW) is accessed. This is followed by a PIN based caller authentication. Finally, a destination telephone number is entered. If the GWs have enough digital signal processing channels and processing capacity, and the backbone (transport) network can support one T1 CAS port's worth of calls, we should be able to simultaneously start 24 voice connection attempts. The call originating GW should be able to process all the 24 connection requests. However, it appears that most of the currently available IP-PSTN GWs can not handle all 24 simultaneous connection requests. Therefore it is necessary to develop a method to determine the number of calls that can be started simultaneously. It is also required to determine the amount of inter call-burst time gap (in millisec. or sec.) so that all 24 calls will be processed using the existing hardware and software configuration and capacity of the GW. In this paper we develop techniques to perform both of the above functions. These are implemented using Hammer visual basic language for testing some commercially available IP telephony GWs.
Usually a voice call is established through multiple stages in IP telephony. In the first stage, a phone number is dialed to reach a near-end or call-originating IP-telephony gateway. The next stages involve user identification through delivering an m-digit user-id to the authentication and/or billing server, and then user authentication by using an n- digit PIN. After that, the caller is allowed (last stage dial tone is provided) to dial a destination phone number provided that authentication is successful. In this paper, we present a very flexible method for measuring call progress time in IP telephony. The proposed technique can be used to measure the system response time at every stage. It is flexible, so that it can be easily modified to include new `tone' or a set of tones, or `voice begin' can be used in every stage to detect the system's response. The proposed method has been implemented using scripts written in Hammer visual basic language for testing with a few commercially available IP telephony gateways.
This paper presents a new architecture and techniques for media-based telephony over wireless/wireline IP networks, called `Beethoven'. The platform supports complex media transport and mobile conferencing for multi-user environments having a non-uniform access. New techniques are presented to provide advanced multimedia call management over different media types and their presentation. The routing and distribution of the media is rendered over the standards based protocol. Our approach offers a generic, distributed and object-oriented solution having interfaces, where signal processing and unified messaging algorithms are embedded as instances of core classes. The platform services are divided into `basic communication', `conferencing' and `media session'. The basic communication form platform core services and supports access from scalable user interface to network end-points. Conferencing services take care of media filter adaptation, conversion, error resiliency, multi-party connection and event signaling, while the media session services offer resources for application-level communication between the terminals. The platform allows flexible attachment of any number of plug-in modules, and thus we use it as a test bench for multiparty/multi-point conferencing and as an evaluation bench for signal coding algorithms. In tests, our architecture showed the ability to easily be scaled from simple voice terminal to complex multi-user conference sharing virtual data.
Interest in the deployment of Wavelength Division Multiplexing in regional and metropolitan networks has increased recently. We consider an example of regional/metropolitan rings and investigate the interaction of node-induced crosstalk with fiber nonlinearities in it. The phenomenon is studied in two cases, namely the case of Non-Zero Dispersion Shifted Fiber operating in the anomalous dispersion regime and the case of Single Mode Fiber with uniformly distributed segments of Dispersion Compensating Fiber. The dependence of the effect of the crosstalk/nonlinear interaction on the frequency difference between signal and crosstalk carriers and on signal power is examined in detail. It is shown that the node-induced crosstalk can interact with fiber nonlinearities and introduce limitations on transmission performance. This interaction should therefore be taken into consideration in designing regional and metropolitan networks.
This paper is focused on the role of Differentiated Services at the access networks in supporting real-time services. Differentiated Services were originally introduced to create virtual partitions in long-haul networks where real-time applications can be transported without being disturbed by other types of traffic. We demonstrate using simulations that Differentiated Services can also be used in broadband access networks in supporting real-time applications. Use of Differentiated Services reduce the amount of state that networks need to keep track of, thus facilitates large-scale deployment of integrated services broadband access networks.