Protocolo activo para transmisiones garantizadas sobre una arquitectura distribuída y multiagente en redes ATM

Abstract

In this doctoral thesis, TAP (Trusted and Active Protocol PDU Transfer) is presented as an innovative architecture for ATM networks due to its active multi-agent and distributive characteristics. The protocol proposed for this architecture offers guaranteed transfer of a privileged group of VPI/VCI connections. We also propose an extension of the AAL-5 layer of ATM which we have called EAAL-5 ( Extended AAL type 5) which is used for the management of privileged end-to end connections. TAP offers Guarantee of Service (GoS) when the network is losing ATM cells and it takes advantage of the inactive periods in the links in order to retransmit the CPCS-PDU-EAAL-5. The protocol we propose uses NACK mechanisms (using backwards RM cells) and is supported by active ATM switches that have a PDU storage memory called DMTE (Dynamic Memory to store Trusted Native EAAL-5 PDU). Our proposed active architecture is based on a MAS (Multi-Agent System) formed by programmable collaborative agents, distributed in the network. The simulations carried out have proved the effectiveness of the PDU recovery mechanism that we have proposed for a better goodput in the net. The TAP architecture is supported over active ATM switches which we have called AcTMs (Active Asynchronous Transfer Mode Switch) and which we have designed with software technology in order to: · guarantee the fair management of entry queues based on WFQ (Weighted Fair Queueing); · manage the control of buffer congestion, using a device inspired by EPD (Early Packet Discard) · and prevent, with VCmerge, the mixing of PDUs coming from different connections. We therefore propose this software technology in order to: fairly distribute the load on the switches; optimize the PDU retransmissions; alleviate the implosion on the sources; prevent PDU fragmentation and decrease the interleaving of cells, thereby optimizing goodput. The AcTM switches also need the appropriate hardware in order to support TAP. Therefore, we propose not only the buffer but also DMTE memory and a set of I/O tables that go with each AcTM port. It has been proved that these hardware requirements are realistic and viable and can be integrated in the active switches. We would also like to point out the multidisciplinary nature of this thesis in which the basis of the research is ATM protocol engineering, complemented by the innovative advantages that the software agents can provide. Nevertheless, the switches that we have finally managed to obtain could be regarded as within the field of specialized architectures; thus several modules of the prototype we have presented could be implemented as hardware components in order to optimize performance. Once the limitations of ATM technology in supporting guaranteed transfer have been identified, which is our principal objective, we describe the general motive for this research in environments where ATM is the basis of IP traffic. An NS (Network Simulator) has thus been used in order to study scenarios in which TAP protocol can significantly improve the already familiar TCP protocol. In order to study the performance of all these proposed improvements, we have used a TAP simulator, which has the advantages that Java language offers for the development of MAS and communication protocols. This simulator can define a variety of scenarios and analyse the results of the simulation of the prototype thus reaching a series of interesting conclusions. The simulations, via ON/OFF sources, analyse point-to-point and point-to-multipoint connections using classes, objects, threads, synchronizations and distributed processes carried out under Java. This dissertation has been organized in three parts in order to adequately structure the contents presented. Part I analyses the research related to this work; thus the first seven chapters describe the basic aspects of ATM technology. Each chapter briefly presents our contributions, which are studied in greater detail in Parts II and III. Thus, in Chapter 1 we point out the basic concepts of this technology and, in Chapter 2, we describe a taxonomy of architectures and protocols for ATM networks which will serve to identify our TAP proposal whose basic architecture is included at the end of the chapter. Chapter 3 deals with the concepts of reliability and GoS, with an emphasis on the latter since this is one of our proposals for the improvement of the general parameters of Quality of Service (QoS). Thus, we explain the mechanism, which is used in order to provide GoS to privileged sources. Following this, Chapter 4, deals with congestion control and fairness, which are applied to ATM switch entry queues. These concepts are also two basic aspects of our proposal to provide solutions to the problem of congestion at privileged sources, while -at the same time- guaranteeing fairness to those sources, which are not privileged. After studying the proposals that have already been published in this area of research, we present an outline of our QPWFQ (Queue PDU Weighted Fair Queueing) algorithm. In Chapter 5 we study the various congestion control mechanisms applied to switch buffers (that already exist) and, after analysing the most common proposals and solutions, we describe our EPDR (Early Packet Discard and Relay) algorithm inspired by EPD and which aims to attend the retransmission requests of congested PDUs. Chapter 6 reviews the literature on software agents designed for communication networks since we intend to explain the relevance of the MAS that we propose as a support for TAP. Thus, our objective is to obtain an active network formed by AcTMs switches whose architecture is outlined at the end of the chapter. In Chapter 7 we justify the distributive characteristics of the TAP protocol over a VPN (Virtual Private Network) formed by AcTMs nodes which exist side by side with non-active switches in the same network. Thus, in Part I we intend to justify our proposals by basing them on the fundamentals of the present technology. Part II describes the general motivation of this thesis, beginning with an analysis of the present limitations of ATM technology, which we propose to solve by using TAP. This part is divided into two chapters: Chapter 8 describes our general objectives whereby the control of congestion at the network nodes not only benefits native ATM traffic but can also be of use for such widespread protocols as TCP. We thus point out the advantages that TAP can provide for present-day networks. In Chapter 9 the limitations of ATM are discussed vis a vis the GoS parameter we have proposed and we explain how TAP can avoid such undesirable problems as PDU fragmentation, traffic interleaving, end-to-end retransmissions and the implosion of traffic sources. The objective of Part III is a comprehensive description of the solutions that we propose and, in the following four chapters, we provide a detailed description both of the architecture and of the protocol that goes with it. Chapter 10 describes the TAP distributed, multi-agent architecture, relating it to the ATM architectural model and analysing each of the hardware and software components of the AcTMs switches. Chapter 11 specifically focuses on the exhaustive description of the complete set of algorithms that make up the TAP protocol and therefore within the MAS which constitutes it. Moreover, the intuitive idea of making use of the periods of inactivity in the network in order to deal with the retransmissions of congested PDUs is formalized as a theory. In Chapter 12 we present the details of the implementation of the TAP simulator proposed as a prototype in order to analyse the results obtained under different scenarios. We argue in favour of the choice of Java language as a tool for the development of protocols and MAS and we then describe the methodology and the most important decisions concerning design as well as the most outstanding classes of Java used in the prototype. This chapter ends with an analysis of the most significant results of the simulations. Finally, Chapter 13 focuses on identifying future lines of action that would give continuity to the whole range of research which has been the object of study of this doctoral thesis.