Université de Liège Faculté des Sciences Appliquées Systèmes et Automatique Esprit Project 5341 Institut d'electricité Montefiore, B28 Université de Liège au Sart Tilman B-4000 Liège 1 (Belgique) OSI 95 The OSI 95 Transport Service with Multimedia support *** The OSI95 Project The OSI95 Project,The OSI95 Transport Service with Nultimedia Support, A. Danthine, University of liège, Belgium (ed.), Research Reports ESPRIT - Project 5341 - OSI95 - Volume 1, Springer-Verlag, 1994, pp. 13-18. Date: July 1994 R94-03
The OSI95 Project OSI 95 is the acronym of ESPRIT II Project 5341 High Performance OSI Protocols with Multimedia Support on HSLANs and B-ISDN. After the presentation of the consortium, the various tasks of this project are described in the order the contributions will be presented in this book. Keywords : OSI95, ESPRIT 1 Introduction It is the changing networking environment discussed in the previous paper [Dan 93] which was at the origin of the ESPRIT II Project, OSI95, started in October 1990 with the initial idea of 1995 as the deadline for its total achievement. The project ended up in February 1993 and it is this first phase that will be presented in detail here. The consortium was headed by BULL(France) and involved, in alphabetical order, Alcatel Bell (Belgium), Alcatel ELIN (Austria), INRIA (France), Institut National des Télécommunications (France), Intracom (Greece), Olivetti Research Laboratories (UK), as well as the universities of Lancaster (UK), Liège (Belgium) and Madrid (Spain). The technical direction of the Project was assumed by the University of Liège. The goal and the central task of this project was the design and the formal specification of a new transport service and of a new transport protocol. This activity was supported by two other main tasks. The first one was the definition of the specific requirements coming from the distributed multimedia applications and ODP systems which may have an impact on the new transport service. The second one was related to the data link layer services provided by ATM networks, in order to identify new underlying network service facilities. Three other tasks were also included in the project. One was related to an in-depth study of the XTP protocol introduced in 1989 and in development at the start of OSI95 with the ultimate goal of producing a chipset. The second task was related to the study of the congestion control, a important mechanism of the transport protocol, and to a preliminary study of the VLSI implementation of this protocol. The third one was related to an extension of the formal specification language LOTOS to handle the time related aspects of the protocol. Last but not least, the project included a task aiming at the standardisation of its results by the creation of a new work item in ISO. We will now describe in more detail the different tasks and goals, not in the order mentioned in this introduction but in the order the contributions will be presented in this book.
14 2. The XTP Study Even if the intention was to work within the framework and spirit of OSI, which the partners feel will accommodate the new developments it was essential to take full account of all relevant research that has taken place outside of OSI, such as the work on new transport protocols. This involves analysing the features of recent transport protocols, particularly the currently proposed XTP (express Transfer Protocol) protocol. The XTP protocol has been derived from the French Military real time Lan Standard GAMT 103. It was felt that it might become a de facto standard of highspeed transport protocol. It is worth mentioning that XTP does not contain any definition of the service it is intended to provide, and does not seem, at the time, to be concerned with the B-ISDN environment. Furthermore, XTP has been neither validated for its logical coherence nor fully evaluated with respect to performance measures. The XTP study included its formal specification in a standardised ISO language, its validation, an evaluation of its performance, and finally an implementation on UNIX. In order to validate the logical correctness of the XTP protocol, its description had to be formally expressed. Due to BULL and INT 1 experience and tools availability, the Estelle specification language was chosen. Finally a prototype implementation of the XTP for demonstration purpose and for an initial performance evaluation was done by INRIA. The machine used for the porting was a UNIX-based commercially available workstation. 3. The Requirements of the New Applications The characteristics of communications in today's environment differ from those for which the initial OSI protocols were designed. Originally the dominant factor was related to terminal/mainframe conversations and file transfer operations. In the case of terminal traffic, messages tended to be short and multiple transactions would be entered within a session that could be maintained for a long period. File transfer in addition implied that a connection between correspondents would persist for a relatively long time. Today's workstations, on the other hand, may operate for a significant length of time on a "large block" of data within their memory and only require communications service intermittently. However, there is a requirement for fast response and high transfer speed. New distributed systems, as implemented in ESPRIT projects, such as COMANDOS or ISA, or as studied in the standardisation bodies under the ODP work item, imply a new pattern of communications between the different parts of integrated distributed applications. Also, support environments for ODP must support distribution transparencies which include, location, concurrency, replication and migration transparency. Other functions of SE-ODP which need support are traders, 1 INT is the acronym of the "Institut National des Télécommunications "(France)
The OSI95 Project 15 security and communications support. Investigation is needed to avoid duplication of functions in the SE-ODP and communications systems. Many new applications will require the handling of multimedia objects composed of voice, text, image and video. Their multimedia nature implies that several datastreams must be handled in parallel; since these streams belong to a single application and are therefore not independent, ways of synchronising these streams must be found. Because of the requirement for multiple data streams, multiconnection facilities are necessary, for example to set up connections for voice and video simultaneously. Since the new applications will often involve more than two partners, protocols linking several entities in a network are required. None of these facilities is currently defined in the OSI standards. The quality of service required by the applications, and the method of mapping this on to the lower-layer protocols, is extremely important. The interaction between these types of protocols and the session and presentation layers is an area requiring investigation. Another protocol requirement is to support the dynamic creation of multiple parallel streams operating at different speeds, thereby giving the application the ability to shape the network to its needs. The automation of these facilities from an application and the optimisation of their use within, for example, a call-based charging regime also requires study. The main aim was to look at these new requirements and discover how they affect the supporting protocols, and to propose what applications services and protocols will be necessary to meet these requirements. The work which involved the University of Lancaster, Alcatel Austria ELIN, Olivetti Research, DIT 2, INRIA and BULL consisted of two parts; one looked at the multimedia requirements, and the other dealt with the requirements of distributed applications, particularly the requirements of SE- ODP. 4. The OSI95 Transport Service Design The basic goal of ULg was to design and specify a transport service and a transport protocol which would not jeopardise the potential bandwidth offered by the new environment created by the high speed LANs ( 100 Mbps). The transport service was intended to be upward compatible with the standard connection-mode transport service, and the underlying service was the standard LLC type 1 service. The protocol had to be designed assuming a LLC access rate 50 Mbps. This implies that the layer 2 service is offered through a high performance controller with a chipset and/or ASIC or VLSI chips. The design of transport protocol had also to take account of the LLC service offered by the B-ISDN through its AALs (ATM Adaptation Layer). One of the first step of the project of the new design was the detailed analysis by ULg of existing transport services and protocols (TP 4 and TCP/IP) to put into evidence all their drawbacks in the new environment, followed by a detailed analysis of XTP, to clearly identify the area where it contributes to the improvement of these drawbacks. 2 DIT is the acronym of "Departamento de Ingenieria de Sistemas Telematicos" (Spain)
16 Protocols imply mechanisms, syntax and implementation. Some people claim that implementation is very often responsible for poor performance and it is true that a poor implementation gives rise to poor performance. But even good implementations are not able to correct the influence of poorly designed mechanisms and to deal with badly designed PDU syntaxes. Good implementations imply that the consequence of the scheduling, of buffers copying and other operating system related issues, be minimised for the performance of the station access rate. Furthermore, the design of a transport protocol must take account of the possibility of implementing it on silicon. The goal of the project was not only to specify new transport service and protocol. It was also to apply LOTOS, a formal description technique, from the very beginning of the design process. The methodology for the design of the service and the protocol will be based on the constraint-oriented style in order to allow an incremental development of the protocol. This method, applicable to the design of large specifications, allows the addition of, the removal of or a modification in a mechanism without jeopardising the other part of the work. It was expected that this basic use of a formal description technique during the process of specification would end up in better protocol. The new methodology has been tested first on TP4 before being used for the OSI95 transport service. The design of a complex LOTOS specifications required the use of adequate tools. The OSI95 project closely followed the work of ESPRIT/LOTOSPHERE and used their LOTOS toolset, LITE, as soon as they were made available to other ESPRIT projects. 5. ATM and the New Communication Environment The ATM style of communication using short, fixed length cells, has gained widespread popularity. This has been enhanced by the CCITT I.121 recommendation of an ATM methodology and cell format for B-ISDN. It need hardly be repeated that ATM is suitable for both fixed and variable rate traffic and can support data and realtime traffic with equal ease. One task, by Alcatel Bell, was to evaluate how these new communications techniques could be incorporated into the OSI data-link layer, particularly when realtime traffic was also considered. This task assessed the service offered by B-ISDN ATM networks; more particularly, it looked at the service of the different types of ATM Adaptation Layer and evaluated how these services would support OSI LLC types of service. The new features of ATM based networks (e.g. flexible bandwidth) could lead to the definition of new types of LLC in order to allow higher layers to exploit fully the capabilities of these ATM based networks. Olivetti has extensive practical experience of slotted ring systems. Such short cell networks are compatible with ATM techniques and Olivetti developed network interface and device drivers to connect ATM networks into standard protocol stacks, including the Internet Protocol and Ultrix V4.2 socket interface. These hardware platforms could later be used for testing the transport protocol over ATM networks.
The OSI95 Project 17 6. Specific Studies The objective of the project was to define a high-speed transport protocol, TPX, optimised for high speed LANs. However, when LANs are interconnected, some users may decide to use this protocol over the interconnection. The purpose of the task was to study suitable congestion avoidance mechanisms that would be usable for high speed networks. The study included a survey of existing algorithms (slow-start, DEC-bit) and the simulation of rate-based control mechanism developed at INRIA. Intracom identified the protocol VLSI implementation problems and provided guidelines for achieving a VLSI implementation of transport protocols. It was expected that the siliconability of a transport protocol, which was investigated in this workpackage, would provide some guidelines for the design of TPX. 7. Time Extension of LOTOS It was the intent of the project to specify as much as possible the new transport service and protocol in the ISO language, LOTOS. It was however anticipated that the new communication and application environments would, in some specific cases, require some extensions to the language. In particular, formal specification of new transport services influenced by the availability of some ATM-based services could require extensions of LOTOS covering the modelling of isochronous as well as asynchronous systems. An objective of this task was to define a model of an extended LOTOS language which could describe the time evolution of systems in a quantitative way, aside to modelling them in terms of events occurrence ordering only, as Standard LOTOS does. The timed LOTOS language model had to be able to describe both synchronous and asynchronous systems. Another language extension was also developed to provide the facility of describing systems behaviour in a probabilistic way; thus allowing the reduction of the nondeterminism of the Standard LOTOS Language when statistical information about the system being modelled is suitable and available. The two extensions above, done by DIT, would make the LOTOS language suitable for analysing systems from the logical point of view, and from the performance point of view as well. The DIT Timed LOTOS proposal was assessed on a well defined case study defined by ULg. The DIT Extended LOTOS model (time + probabilities) was validated through modelling case studies selected from high speed networking technologies, services and protocols (XTP, FDDI, ATM,...).
18 8. Standardization The interest of developing a new Transport service and a new transport protocol would have been limited, if this effort has not been integrated in the standardization activities. This was the reason why one of the goal of the project, strongly supported by the CEC, was the creation of one or several New Work Items (NWI) on High Speed Protocols and Services in various standardisation environments such as ISO/IEC JTC1/SC6, ETSI/NA5, to avoid the repetition of the FDDI standardisation which has been carried out entirely in the US and has given a competitive edge to US companies for the development of FDDI-based products. The creation of such a NWI was a prerequisite to the promotion of new transport service and protocol in the Standardisation Organisations. The creation of such a NWI was clearly a success criterion for the OSI95 project. 9. Conclusion In this book, we will present, following the sections of this paper, the main results of the first phase of the OSI95 project.