Skip to main content

Über dieses Buch

The main objective of this workshop was to review and discuss the state of the art and the latest advances· in the area of 1-10 Gbit/s throughput for local and metropolitan area networks. The first generation of local area networks had throughputs in the range 1-20 Mbit/s. Well-known examples of this first generation networks are the Ethernet and the Token Ring. The second generation of networks allowed throughputs in the range 100-200 Mbit/s. Representatives of this generation are the FDDI double ring and the DQDB (IEEE 802.6) networks. The third generation networks will have throughputs in the range 1-10 Gbit/s. The rapid development and deployment of fiber optics worldwide, as well as the projected emergence of a market for broadband services, have given rise to the development of broadband ISDN standards. Currently, the Asynchronous Transfer Mode (ATM) appears to be a viable solution to broadband networks. The possibility of all-optical networks in the future is being examined. This would allow the tapping of approximately 50 terahertz or so available in the lightwave range of the frequency spectrum. It is envisaged that using such a high-speed network it will be feasible to distribute high-quality video to the home, to carry out rapid retrieval of radiological and other scientific images, and to enable multi-media conferencing between various parties.



Architecture of High-Speed LANs and MANs

Network Management and Control in Broadband Telecommunication Networks: Research Issues and Some Partial Solutions

In this paper we first survey briefly the state of the art in current telecommunication networks. We then focus on ATM-based broadband integrated networks of the near future. Examples include the IEEE 802.6 DQDB metropolitan area network standard and Bellcore’s Switched Multimegabit Data Service (SMDS), which uses the 802.6 protocol and ATM at its lowest layers. Looking further into the future it is projected that lightwave- (fiber optic-) based networks may be deployed running at Gbps and Tbps rates. These very high speed networks introduce critical problems of management and control. These include the design of end-to-end lean (lightweight) protocols; access, flow and congestion control; and the real-time detection and identification of network faults when they occur. Some representative ongoing work at Columbia in each of these areas is described, with a hope that it may stimulate additional work elsewhere.
Mischa Schwartz

The Medium Access Control

Wide Area Connectionless IEEE 802.6 MAN Services

Development of connectionless data services based on the IEEE 802.6 MAN standard is discussed. An architecture of a public network capable of providing such services is outlined. It offers the subscriber an interface that conforms to the 802.6 standard and that is supported by broadband connectionless packet switching. In the core, switching is by connectionless routers over a virtual network on an ATM fabric. At the periphery it has connectionless shared medium subnetworks that perform end switching as well as traffic concentration. In evolution, the network will grow from subscriber interface and periphery to core. Its development can form a path for B-ISDN.
Z. L. Budrikis

New Access Schemes for Gbit/s LANs and MANs

Distributed-Queue Multiple-Access (DQMA) and Cyclic-Reservation Multiple-Access (CRMA) are new access schemes for Gbit/s local and metropolitan area networks. These schemes are based on slotted unidirectional bus structure, both folded and dual-bus configurations. Like the DQDB, both DQMA and CRMA access schemes achieve full throughput independent of network speed and distance. However, they have two significant advantages over DQDB. First, they provide throughput fairness even at high speeds and large distances. Secondly, by allowing reservation of multiple consecutive slots, they make segment labeling unnecessary, and facilitate packet reassembly significantly.
Hans R. Müller, M. Mehdi Nassehi, Johnny W. Wong, Erwin Zurfluh, Werner Bux, Pitro Zafiropulo

Cell Switching and Access Schemes for HS-LANs and HS-MANs

The third generation of LANs should reach the Gbit/s range. Moreover, the integration of different types of traffic on the same medium will impose constraints in terms of throughput, delay, delay dispersion, reliability and sequenced delivery. The third generation should take all of these constraints into account. In the same time, high-speed telecommunication technology is emerging through Broadband ISDNs (Integrated Services Digital Networks). The standardization of B-ISDN follows the ATM principle (Asynchronous Transfer Mode) chosen by the CCITT Study Group XVIII.
The purpose of this paper is to give some ideas of expected evolution in the field of High Speed LANs and MANs. As the present proposals do not seem to be the best solutions for the Gbit/s range, we describe a new access technique for LANs or MANs, compatible with the ATM technique.
Guy Pujolle

First Experiences in High-Speed LANs

High Speed Local Area Networks — What, why, when and how: Planning, installation and first experiences of a HSLAN in an heterogeneous environment

The paper begins with a description of characteristics of High Speed Local Area Networks. Following that the pro’s and con’s of the main standardisation candidates (FDDI, DQDB) are presented together with a comparison of their relative merits. A major part of the manuscript is devoted to the detailed description of lessons learned from a FDDI installation within a heterogeneous environment (the network is to be used as a backbone for the interconnection of university institutes which have different applications as well as different computing facilities). The paper ends with some perspectives and a few speculations for the future of HSLAN development, market potential and acceptance.
Otto Spaniol

ONERA Gbit/s LAN: first experiment and use

ONERA has installed the first elements of an UltraNet® Gbit/s LAN in December 89. The first purpose of this LAN was to connect a CRAY XMP416 Supercomputer to workstations and a Graphical Display (Frame Buffer).
Thus, two types of experiments were carried on, i.e.
use of the Frame Buffer to display images directly from the CRAY;
Transfers of files between the CRAY and workstations.
The Graphics experiments were carried to evaluate:
  • the performances of the network for images transfers.
  • the benefit that a user can draw from this device by monitoring computations on the CRAY, or by creating an animation film from a program running on the CRAY.
  • the effort needed to realize a Graphical Interface between the user program and the transfer unities.
  • the resources needed on the Supercomputer to fully take advantage of this device.
The Networks experiments consisted of measuring data transfer between the CRAY and workstations, using both the UltraNet® as a backbone and figures compared with these of NASA AMES (1st UltraNet® user) and Stuttgart RUS (1st european site).
The odd nature of the first usage of this broadland LAN has to be underlined: visual, on the fly analysis of voluminous data being processed on the supercomputer loads the network up to quasi saturation for a single user.
It is planned however, to use this powerful tool in a more distributed fashion, for a variety of applications: supercomputer — file server relations, supercomputer — superworkstations cooperative processing, FDDI LANS federation.
Interfacing ONERA UltraNet® Gbit/s LAN with the upcoming 100 Mbit/s Paris Area R&D MAN is under study.
Most of these projects should reach maturity within next 3 to 4 years. In the meantime, both supercomputers and workstations will have gained a factor of 10 in computing power, which will impact critical network segments. Ethernets will have to upgraded to FDDI LANs and FDDI backbones to Gbit LANs.
On going experiments in ONERA are regarded as necessary to efficiently face the rapidly evolving demand of its computer users both from research and industry in the national and european aeronautical segment of economy.
J. P. Peltier, J. Zeyons

LION and MAX, the experiences of two ESPRIT Projects on High-Speed MANs

After introducing the necessary definition of LANs, MANs, WANs, two European projects are presented: LION (a LAN) and MAX (a MAN). These are integrated networks, so two flow types run on such networks:
  • circuit level or synchronous communications,
  • packet data level or asynchronous communications.
We present LION project results and explain why making an efficiently packet level is difficult via Hardware and Sofware aspects. Besides, we introduce what is the consortium current research.
PH. Legay, G. Roullet

Metropolitan Area Networks

Fault Tolerance and Circuit Service in Deflection Networks

Deflection Networks are very attractive communications systems, since they are capable to extend to MANs and wide area networks the services provided by LANs. These networks are based on Mesh Topologies and use the Deflection Routing. Unlike ring and bus topologies, mesh networks provide multiple paths between sources and destinations and offer very high connection reliability and traffic-handling capability. If the network topology changes because of node and link failures, a further mechanism is needed to dynamically estimate the shortest-path distances required to guarantee the proper operation of the Deflection Algorithm. The aim of this paper is twofold. First we show the fault-tolerance capabilities of this architecture when the Backward Learning mechanism is used to update routing tables and second we present a distributed mechanism able to provide circuit services with guaranteed bandwidth.
Flaminio Borgonovo, Luigi Fratta

A Linear Lightwave MAN Architecture

A new type of architecture for an optical Metropolitan Area Network is described, based on establishing controllable, optically transparent paths among network users. The networks in question perform only linear operations on optical signals (power combining, splitting and linear amplification), essentially operating as a dynamically controlled “ether” within which light beams are selectively directed between prescribed source-destination pairs on demand. Because they are controllable, they are capable of being reconfigured in response to changing load conditions or component failures.
The LLMAN architecture is realized by superimposing some simple multiple access procedures on a linear optical network structure. At the optical level, point-to-point or multipoint connections are established by controlling the optical parameters at the network nodes. These connections create “multicast/multiaccess subnets” within which logical connections are established through multiple access techniques. The MAN’s described here should be capable of supporting a large number of user stations (for example, 1,000) operating at user speeds of the order of a Gigabit/sec, and using arbitrary mesh-type topologies spread over geographical areas that may extend to diameters of several hundred kilometers without regeneration.
Thomas E. Stern

Routing in Large Metropolitan Area Networks Based on Wavelength-Division Multiplexing Technology

High-speed networks require fast routine procedures and cost-effective networks require uncomplicated station design. The use of detour routing in the Wavelength-Division Optical Network (WON) allows hardware-implemented source routing and stations with reduced buffering requirements. We describe detour routing and report results of experiments that compare specially optimized WONs to the Manhattan Street Network (MSN). The optimized WONs, found via simulated annealing, achieve higher performance than the MSN. We also examine the effect on performance of introducing additional transmission buffers into the station.
Joseph A. Bannister, Luigi Fratta, Mario Gerla

Impact of High Performance and Integrity Requirements on Metropolitan Area Network’s Architecture and Technology

Considerable progress has been made on the development, implementation and definition of standards for Metropolitan area networks (MAN’s) service provision. Concepts have changed quite dramatically since inception and a number of unresolved aspects remain. Some of the background will be described, since they have had an influence on the performance characteristics.
Patrick F. Baker


Problems Arising from Deflection Routing: Live-lock, Lockout, Congestion and Message Reassembly

Deflection routing is a technique that has been proposed for slotted, metropolitan area networks. When there are no buffers available at an intermediate node in the network, packets of data are forced to take longer paths through the network. Because this technique only uses local information at a node, it is possible to apply it to high rate networks in which the propagation delay on a link is much greater than the time it takes to transmit a packet.
Several problems encountered with this technique are described in this paper, including live-lock, source lockout, congestion spreading, and reassembly buffer overflow. Live-lock and congestion spreading are unique to deflection routing, source lockout occurs in many slotted networks, and reassembly buffer overflow is common in datagram networks. It is shown that random deflection rules eliminate live-locks, and that the other problems can be managed by end-to-end mechanisms when they persist.
N. F. Maxemchuk

Towards Formal Structures for IEEE 802.6

This paper explores the use of a methodology inspired by “Formal Definition Techniques”, or FDTs, for the description of DQDB, Distributed Queue — Dual Bus, a draft standard for an access procedure to be used as a primary mode of transmission for Metropolitan Area Networks. The thesis here is that these “formal structures” are pimarily useful for delineating, discovering and possibly avoiding “extreme” behavior. Extreme behaviour can manifest itself in malfunctioning software, specification incompleteness, logical inconsistencies, and/or unexpected or unforseen actions. Statistical modelling is important for determining expected behaviour but its validity may be compromised if the system does not behave as expected. This paper concentrates on the the small but important part of the access mechanism, the DQDB layer of the draft standard rather than other aspects of the protocol. It is a preliminary analysis that points out some of the power and problems of the approach.
Michael J. Ferguson

Modelling of the DQDB Access Protocol and Closed-Form Approximation

We present an analytic performance study of the DQDB medium access protocol. The major subject of the queueing analysis is to derive closed-form solutions, which should be easy to evaluate but deliver sufficiently accurate performance measures to study protocol behaviors. We use a decomposition of the medium access delay by means of the technique of embedded modelling. The non-isochronous station-to-station traffic matrix can be chosen arbitrarily and an amount of isochronous traffic is taken into account. Two cases are considered: i) Poisson input traffic with a continuous-time model and ii) general discrete-time traffic in conjunction with a discrete-time modeling environment. The accuracy of the approximation technique developed in this paper is appropriate for a wide range of protocol parameters.
P. Tran-Gia, Th. Stock

Approximate Performance Model and Fairness Condition of the DQDB Protocol

The Distributed Queue Dual Bus (DQDB) proposal is the emerging candidate for the Metropolitan Area Network standard of the IEEE 802.6 committee. General properties of the DQDB access scheme such as fairness or priority issues are now well known. Therefore, different versions of the protocol have been released to address the existing problems. We develop a simple probabilistic model of the DQDB protocol under saturated condition in order to avoid excessive simulation runs. The model is based on a two step analysis taking into account the initial utilization state of the network. The accuracy of the model has been checked by comparaison with simulation and found to be very accurate. Moreover, we derive the fairness condition of the DQDB protocol and discuss some policies to achieve it.
Serge Fdida, Harry Santoso

An Analytical Model for the High Speed Protocol DQDB

We give an analytical evaluation of the high speed protocol, DQDB, under a simple model. We suppose that the number of connected nodes is (virtually) infinite, providing a global load λ < 1; the distances between nodes are large and randomly distributed. An analogy is developed with queueing analysis, which leads to analytical evaluation via closed formulas.
Philippe Jacquet, Paul Mühlethaler

Frame Delay Analysis of the DQDB Protocol

An approximate delay analysis for the non-isochronous traffic in DQDB networks is provided. Modeling the network stations as M/G/1 queueing systems with batch arrivals, and utilizing results from previous analysis for segment delay, closed- form expressions for the average frame delay at individual stations are obtained. The results are then compared with simulations, and it is shown that, for ligth to moderate bus loads, the accuracy of the approximation is kept within acceptable levels.
Luís F. M. de Moraes

Performance of ATM Techniques

Fast Packet Switch Architectures and the Tandem Banyan Switching Fabric

We first present a brief review of the various fast packet switch architectures proposed so far. We then introduce a new space-division switch architecture based on banyan interconnection networks, called the Tandem Banyan Switching Fabric. It consists of placing banyan networks in tandem, offering multiple paths from each input to each output, overcoming in a very simple way the effect of blocking in banyan networks. It is shown to be simpler and to perform better than the Batcher-banyan and buffered- banyan switching fabrics, and to perform as well as space-division architectures offering N 2 disjoint paths (such as the Knockout switching fabric) while remaining considerably simpler.
Fouad A. Tobagi, Timothy Kwok

An Approximation Analysis of a Shared Buffer ATM Switch Architecture under Bursty Arrivals

We present an approximation algorithm for the performance analysis of a shared buffer ATM switch architecture. The arrival process to each input port of the ATM switch is assumed to be bursty and it is modelled by an Interrupted Poisson Process. Comparisons against simulation showed that the approximation algorithm has a good error-level.
H. Yamashita, H. G. Perros, S.-W. Hong

Some Results on Traffic Control and Resource Allocation in ATM networks

We first consider the problem of minimizing the loss in a system of two finite capacity queues in series with no blocking at the first queue. The issue is to choose an optimal ratio of the service rates and an optimal distribution of the buffer space between the two queues. Then we address the same optimization problem for a three queue system with multiplexing at the third queue and we discuss the potential applications of the results obtained for dimensioning broadband ISDN networks based on ATM.
Pierre-Jacques Courtois, Guy Scheys

New Trends in High-Speed Communications

Reliable Broadcast and Agreement with Unreliable Asynchronous Highspeed LANs and MANs

High-speed LANs and MANs are potentially able to handle high traffic loads efficiently. Also, most often, they offer efficient message broadcast services.
These features are particularly appealing to designers of distributed computing systems because many of the algorithms embedded in distributed systems rest on massive utilization of message broadcast.
This paper examines those conditions under which it is possible to achieve reliable message broadcasting over unreliable high-speed LANs and MANs in the presence of crash and omission failures. For the sake of rigor and generality, an asynchronous model of LANs/MANs is considered, with majority voting based agreement utilized to obtain reliable broadcast.
Feasibility conditions are established and expressed as a lower bound on the number of processes needed in a high-speed LAN/MAN that is meant to tolerate any combination of crash and omission failures.
G. Le Lann

Reliable Multicasting in High-speed LANs

There is an increasing number of distributed applications, some of them fault-tolerant, and it has been recognized that its construction may benefit from the existence of reliable broadcast protocols. Reliable broadcasting has deserved considerable attention recently. Some systems are clock-driven, exhibiting tight synchrony: they rely on clock synchronization and space redundancy. Others, like the AMp, an atomic multicast protocol for local area networks, are clock-less.
In application-independent systems, most of the time domain requirements are not of the hard real-time kind — which clock-driven protocols are most suited for — but rather of the on-line, or soft real-time kind. To encourage utilization of reliable broadcast protocols in such applications, it is mandatory that benefits in quality of service are not considered too costly in performance. Clock-less protocols provide an answer to this requirement, since they can trade synchronism for fast termination.
This paper deals with the performance implications of supporting distributed applications, with clock-less reliable broadcast protocols. The question of the network influence in protocol performance is analyzed. We predict the performance of AMp on two target LANs: the l0Mb/s token-bus, and the 100 Mb/s FDDI. We observe that the figures achieved with a LAN based approach are good enough to match the requirements of high performance distributed kernels. Additionally, we show that, under these conditions, using a high-speed LAN does present an advantage: not only throughput increases, a natural consequence, but also speed, measured in duration of single AMp executions, for small messages.
Paulo Veríssimo, Luís Rodrigues

Management Considerations in moving from a 10Mbit LAN to a 100 Mbit LAN

Our university is preparing to introduce a 100Mbit/s campus backbone. The current position has been heavily influenced by the resource mechanism used by central government. It has led to the creation of JANET, a country-wide X25 academic network, and the adoption of the same protocol standards on our 10Mbit/s LAN. A JANET upgrade to 2Mbit/s is in hand and a 3 year local university 100Mbit/s LAN programme has started, with most sites opting for FDDI.
We are installing Ethernets within departments and interfacing them to the Ring 10Mbit/s campus backbone which we have had since 1978. The installation of fibre cabling will begin in Autumn 1990 with a 100Mbit/s backbone available by 1992. The reasons for choosing FDDI over other technologies are discussed. Our first project will be an X25 JANET to FDDI gateway.
By 1994 we expect most of our campus computing service to be based on a small number of large UNIX† machines networked via FDDI, with user access from X terminals via departmental Ethernets.
A. L. Ibbetson, P. W. Riley, E. B. Spratt

Performance Issues

On Overload Control of High Speed Switching Nodes

In this report three basic principles for overload control of high speed telecommunication switches are presented, evaluated and compared. Then the concept of priorities is introduced into these models. Priorities are given to jobs related to calls being in progress. In a second case jobs associated to one out of many different services are prioritierised. The approach taken here is applicable to centralized processor systems as well as to a loosely coupled fully distributed processor environment.
Ulf Körner, Christian Nyberg

Proposal-Petition Protocol: Models for Performance Evaluation

In this work we introduce a significant improvement to the Proposal-Petition Protocol which was defined in [4]. This improvement affects the Petition process by allowing a global three code alphabet search system to be transmitted over the channel. Whenever a petition message is being sent by more than one station simultaneously or no station is sending it, the previous Petition process is used, but if only one station is sending the Petition message, a new alphabet element is sent, so, the global system know wich station will need the token. Our results show that the overhead is drastically reduced if the system is lightly loaded, and tends to be the same as the token-passing scheme when the system is heavily loaded.
Ramon Puigjaner, Josep M. Fuertes


Design, Implementation and Performance Analysis of an ETHERNET to LION Gateway

The introduction of the new High-Speed LANs has not only to take into account the integration of the new services but also to provide powerful interconnections with traditional LANs like the ETHERNET. This paper addresses the ETHERNET internetworking solution adopted for a Local Integrated Optical Network (LION). The architecture of this internetworking solution is described, both in hardware and software terms, and the real time requirements are highlighted, describing the implemented system. Finally, the Gateway performance analysis is given and the selection of the implementation parameters becomes obvious.
T. Antonakopoulos, J. Koutsonikos, V. Makios

Design and Analysis of Flow Control Protocols for Metropolitan Area Networks

In this paper we study the problem of flow control for LAN’s interconnected through a high speed MAN. We consider several input bridges feeding data through the MAN to an output bridge and specifically concern ourselves with the avoidance or minimization of buffer overflow at these bridges. We study the behavior of four flow control policies that differ from each other according to the type of information passed aroundamong the bridges. The most complex protocols use queue length information whereas the simpler protocols use either no information or packet age information. We show that the protocols using queue length information are optimal in the sense that they minimmize buffer overflow for a broad class of systems. In addition we compare the performance of these policies through a combination of analysis and simulation. We observe that using age information, which is relatively inexpensive to acquire, yields half of the benefit of queue length information. Furthermore, if most of the buffers are allocated to the output bridge, then there is little difference between the behavior of these policies. This suggests that simple protocols may work well under such allocations. Last, we study the issue of fairness when input bridges are not identical. We observe that the policies based on queue length information provide fairer treatment when the performance metric is probability of loss and the simple policies provide fairer treatment in the case of mean packet delay.
Don Towsley, Serge Fdida, Harry Santoso

Control in High-Speed Networks

Analysis of an Asymmetric Polling Model with Cycle-time Constraint

Local area networks (LANs) have been extensively constructed to efficiently serve nonreal-time traffic such as data traffic. Due to the multiplexing techniques for multi-media communications, it is required to integrate nonreal-time traffic which tolerates possibly long delay and real-time traffic (e.g.,voice,video) which poses strict delay constraint. However ordinary LAN protocols cannot adequately meet this requirement.
The FDDI protocol employs a token passing scheme with cycle-time constraint, which limits the amount of nonreal-time traffic transmitted by a station to preserve smooth transmission of real-time traffic. With the scheme, real-time traffic is guaranteed to be transmitted within predetermined delay and nonreal-time traffic is transmitted unless its transmission causes real-time traffic excessive delay. Most works related to a token passing scheme with cycle-time constraint have investigated the allocation of bandwidth. Mean waiting time for messages in a symmetric system is analyzed by Takagi, but no asymmetric system has been analyzed yet.
In our study, we consider the mathematical model for a token passing scheme with cycle-time constraint and analyze waiting time for messages in an asymmetric system, where each message arrives according to a Poisson process. Service time and switchover time are constant, so that we can discretize time axis. Each station can transmit a message unless it causes excessive delay of real-time traffic. We consider the case where the load of each station may differ and the restriction on token rotation time at each station may differ. From the numerical results, it turns out that the mean message waiting time depends on the location of the station, and that the saturation point of load at which each queue becomes unstable depends on the selection of target token rotation time.
Kazuta Nakamura, Tetsuya Takine, Yutaka Takahashi, Toshiharu Hasegawa

Waiting Times in Token Rings with Helical Window and Window-Gated Service

We extend our recent analysis of symmetric token rings to include a more practical window-based access rule. Instead of constant-size “helical” windows (which sometimes require the idle token to be artificially delayed), we now handle variable-size “gated” windows, defined as the minimum of a design parameter, w, and the lag of the algorithm when the token arrives. For suitable values of w, we can give exact closed-form expressions for the mean, variance and moment generating function for the system time, assuming Poisson traffic at rate G/N at each of N stations, arbitrarily spaced around the ring, and a general independent service time distribution common to all stations. These restrictions on w ensure that the lag induced by each packet transmission requires the idle token to make an integral number of tours of the ring. Suitable values for w are quite easy to find when the walk time is much smaller than the mean service time, but may not exist when the walk time is much larger. Nevertheless, we show via simulation that the analytical solution can still be a very good approximation even when the conditions for exact results are not met. Window-gated service also has significant performance advantages over more traditional service disciplines, especially when the walk time is small or the variance of the packet length distribution is large.
Mart L. Molle

Priorities in ATM Networks

In this paper we develop an exact model to evaluate the benefits of using space priorities in an ATM network when the traffic source is bursty. The buffer mechanism modelled is the partial buffer sharing. The bursty source chosen is a Markov modulated Poisson process (MMPP).
Jorge García, Olga Casals


Weitere Informationen