In this thesis we develop a two component mixture model to perform a Bayesian regression. We implement our model computationally using the Gibbs sampler algorithm and apply it to a dataset of differences in time measurement between two clocks. The dataset has ``good" time measurements and ``bad" time measurements that were associated with the two components of our mixture model. From our theoretical work we show that latent variables are a useful tool to implement our Bayesian normal mixture model with two components. After applying our model to the data we found that the model reasonably assigned probabilities of occurrence to the two states of the phenomenon of study; it also identified two processes with the same slope, different intercepts and different variances. / McAnulty College and Graduate School of Liberal Arts; / Computational Mathematics / MS; / Thesis;

Optical packet switching (OPS) is the most prominent candidate transport solution that can seamlessly integrate electrical and optical layers by transferring certain switching functionality from electronics to optics, thus alleviating unnecessarily slow and expensive optical-electrical-optical conversions and signal processing at the switching node. Optical Label Switching (OLS) is an important aspect of the optical packet switched network that enables very low-latency forwarding of ultra-high bit-rate, protocol-independent packets entirely in the optical domain. The objective of the proposed research is to develop novel, efficient techniques to realize several key enabling technologies such as optical label generation and encoding, optical label swapping, all-optical buffering, and spectral efficient transmission system for optical label switched based OPS networks. A novel scheme of generating optical label at the ingress node using optical carrier suppression and separation (OCSS) technique is proposed. This scheme does not suffer from any unavoidable interference, limited extinction ratio or strict synchronization requirements between payload and label as observed by the currently available other label generation methods. One of the primary challenges to realize optical label swapping at the core node of scalable OLS network is the insertion of new optical labels without any wavelength conversion for same wavelength packet routing. A novel mechanism to realize same wavelength packet switching without using any conventional wavelength converter in the OLS network carrying differential phase-shift keying (DPSK) modulated payload and on-off keying (OOK) modulated optical label is demonstrated. Also a new dense wavelength division multiplexing (DWDM) optical buffer architecture using optical fiber delay lines that can provide wavelength selective reconfigurable variable delays is proposed. Optical packet switching provides automated, reconfigurable, and faster provision of both wavelength and bandwidth with finer granularity in the optical layer. However, a newer, cost-effective, and spectrally efficient optical transmission technology is essential to support the explosive bandwidth demand expected by the future optical packet switched networks. To meet this challenge, a spectrally efficient solution for transporting 40 Gbps per channel data over 50 GHz spaced DWDM system is developed by exploiting optical carrier suppression and separation technique and optical duobinary modulation.

Este trabalho apresenta uma comparação e avaliação das arquiteturas de roteamento IP e IP sobre MPLS. Inicialmente, são apresentados alguns conceitos básicos dos protocolos IP e MPLS, e suas relações com a camada 2 do modelo OSI. Duas características foram analisadas: o processo de encaminhamento e as granularidades para engenharia de tráfego. A medição de desempenho foi feita com computadores Linux realizando a tarefa de roteamento/encaminhamento de datagramas IP em diferentes tipos de redes Ethernet. Com base nos resultados dos testes, é feita uma discussão sobre o desempenho e topologias adequadas para cada arquitetura. / This work presents a comparison and evaluation of IP and the IP over MPLS routing architectures. Initially, some basic concepts of IP and MPLS protocols and their relationship with layer 2 of OSI model are introduced. Two characteristics were analyzed: the forwarding process and traffic engineering granularity. The performance analysis was realized on Linux computers routing/forwarding IP datagramas at different types of Ethernet networks. Considering the results of the tests, a discussion about performance and suitable topologies is made for each architecture.

Many reports have been written about the techniques behind Virtual Private Networks (VPN) and Multi Protocol Label Switching (MPLS). They usually deal with the low level design of the software implementing a specific technique. The initial products are usually not mature enough to run in a large network or have to be adjusted in some way to fit. This report investigates the different ways of implementing strict layer 2 Virtual Private Networks in an existing nation-wide Gigabit Ethernet. The infrastructure in use, as well as the hardware, has to be used without major changes. Since 1998/1999, when MPLS first started in the laboratories, development has continued. Today it is possible to introduce MPLS or tunneled national virtual local area network into an existing network. This requires high speed, fault tolerant, and stable hardware and software. Going beyond the separation of traffic at layer 3 using Virtual Private Networks, i.e., IPSec, we can tunnel layer 2 traffic through a network. Although the first layer 3 VPN products are already in use, layer 2 VPNs still need to be evaluated and brought into regular use. There are currently two ways of tunneling VLANs in a core network: tunneled VLANs (or as Extreme Networks calls them, VMANs) and MPLS. This project showed that it is possible to start with a VLAN-only solution, and then upgrade to MPLS to solve scalability issues. The VMAN solution can not be used at Arrowhead, since there are too many disadvantages in the way Extreme Networks has implemented it. However, a mix of tunneling VMAN in a VLAN core is possible, and enables customer tagging of VLANs in a Layer 2 VPN. Furthermore, the testing of EAPS and per-VLAN Spanning Tree Protocol turned out well, and showed that EAPS should not be used when there is more than one loop.

Software-Defined Wide Area Networks is a technology that has received a lot of attention in recent years. The technology enables direct Internet access without having to direct traffic through a data center. It also enables the use of less expensive Internet connections to build the Wide-Area Network, expensive Multiprotocol LabelSwitchingconnectionsarenolongernecessary. This results in greater responsibility for enterprises when it comes to ensuring performance within the network, where the previous dedicated links with a guaranteed performance by Internet Service Provider have been replaced with unpredictable best-effort connections. The project targets the challenges Software-Defined Wide Area Networks introduces in terms of monitoring, by examining the required network transparency between Wide-Area Network edges. A selection of Network Management Systems was evaluated with the requirements specification as a basis. The results show that more diligent monitoring at the edges is needed. Visibility into the underlay network, providing transportation, has to be available. Where the enterprises need to be able to examine path-selection of traffic per application. To ensure that the root of problems can be located in case of performance issues. This was not obtained by the evaluated Network Management Systems. Moreover, it turned out that there was no common standard available for monitoring the technology at the time of the project. The lack of knowledge in the field aimed to decrease with the degree project and the result is of greatest interest to companies that have made the transition to Software-Defined Wide Area Networks or are about to do so. / Software-Defined Wide Area Networks är en teknik som har fått mycket uppmärksamhet de senaste åren. Tekniken möjliggör direkt internetåtkomst utan att behöva dirigera trafik genom datacentret. Det möjliggör också användningavbilligareinternetanslutningar för att bygga Wide-AreaNetwork, dyra Multiprotocol Label Switching anslutningar är inte längre nödvändiga. Detta resulterar i ett större ansvar för företag när det gäller att säkerställa prestanda inom nätverket, där de tidigare dedikerade länkarna med garanterad prestanda av internetleverantörer har ersatts med oförutsägbara anslutningar som skickar trafik efter bästa förmåga. Examensarbetet riktar sig mot de utmaningar som Software-Defined Wide Area Networks innebär när det gäller övervakning, genom att undersöka den nödvändiga nätverkstransparensen mellan Wide-Area Network kanterna. Ett urval av nätverkshanteringssystem utvärderades med en definierad kravspecifikation som grund. Resultaten visar att det behövs mer noggrann övervakning i Wide-Area Network kanterna. Insyn i transportlagret måste vara tillgänglig och företagen behöver kunna granska vägval av trafik per applikation, för att säkerställa att ursprungsfel går att lokaliseras i händelse av prestandaproblem. Detta erhölls inte av de utvärderade nätverkshanteringssystemen. Det visade sig dessutom att det inte fanns någon gemensam standard tillgänglig för övervakning av tekniken vid tidpunkten för examensarbetet. Examensarbetet syftar till att minska kunskapsbristen inom området och resultatet är av störst intresse för företag som gjort övergången till Software-Defined Wide Area Networks eller är på väg att göra det.

During the last years, MultiProtocol Label Switching (MPLS) has been deployed by most large Service Providers (SP). The main driver for MPLS deployment is the ability to provide new services with stringent Service Level Agreements (SLAs) such as layer-2 and layer-3 Virtual Private Networks (VPNs) as well as Voice and Video over IP. Most of these services are already deployed inside single SP networks. However, customers now require world-wide VPN and VoIP services. Therefore, SPs need to collaborate to offer these services across multiple SP networks. Inside a single SP network, each node usually knows the complete topology of the network with the load and delay of all the links. Based on this information, each router is able to compute constrained paths toward any other router inside the SP network. Then, it can establish a connection and reserve resources along the computed path with the Resource reSerVation Protocol (RSVP-TE). However, when services with stringent requirements must cross multiple SP networks the computation of the path becomes a problem. Routers in different SP networks exchange routing information by using the Border Gateway Protocol (BGP). BGP provides reachability information. It does not distribute complete topology, delay and bandwidth information. One way to provide guaranteed services crossing different SPs is to delegate the computation of the paths to a Path Computation Element (PCE) that learns the topology of the different SPs. However, this requires that SPs reveal information that they usually consider confidential, their topology. In this thesis, we perform active measurements to show the difficulty to engineer the interdomain traffic with BGP. MPLS together with RSVP-TE provide much more control on the traffic. We define extensions to RSVP-TE for the protection of inter-AS MPLS paths. The aim is to be able to provide the same service guarantees as inside a domain while keeping the internal topology of SPs confidential, as required by SPs. We propose and evaluate distributed techniques relying on PCEs for the computation of interdomain constrained paths respecting the latter confidentiality requirement.

Les algorithmes d'inférence ou d'optimisation possèdent généralement des hyperparamètres qu'il est nécessaire d'ajuster. Nous nous intéressons ici à l'automatisation de cette étape d'ajustement et considérons différentes méthodes qui y parviennent en apprenant en ligne la structure du problème considéré.La première moitié de cette thèse explore l'ajustement des hyperparamètres en apprentissage artificiel. Après avoir présenté et amélioré le cadre générique de l'optimisation séquentielle à base de modèles (SMBO), nous montrons que SMBO s'applique avec succès à l'ajustement des hyperparamètres de réseaux de neurones profonds. Nous proposons ensuite un algorithme collaboratif d'ajustement qui mime la mémoire qu'ont les humains d'expériences passées avec le même algorithme sur d'autres données.La seconde moitié de cette thèse porte sur les algorithmes MCMC adaptatifs, des algorithmes d'échantillonnage qui explorent des distributions de probabilité souvent complexes en ajustant leurs paramètres internes en ligne. Pour motiver leur étude, nous décrivons d'abord l'observatoire Pierre Auger, une expérience de physique des particules dédiée à l'étude des rayons cosmiques. Nous proposons une première partie du modèle génératif d'Auger et introduisons une procédure d'inférence des paramètres individuels de chaque événement d'Auger qui ne requiert que ce premier modèle. Ensuite, nous remarquons que ce modèle est sujet à un problème connu sous le nom de label switching. Après avoir présenté les solutions existantes, nous proposons AMOR, le premier algorithme MCMC adaptatif doté d'un réétiquetage en ligne qui résout le label switching. Nous présentons une étude empirique et des résultats théoriques de consistance d'AMOR, qui mettent en lumière des liens entre le réétiquetage et la quantification vectorielle.

Traffic Engineering (TE) is the stage which deals with geometric design planning and traffic operation of networks, network devices and relationship of routers for the transportation of data. TE is that feature of network engineering which concentrate on problems of performance optimization of operational networks. It involves techniques and application of knowledge to gain performance objectives, which includes movement of data through network, reliability, planning of network capacity and efficient use of network resources. This thesis addresses the problems of traffic engineering and suggests a solution by using the concept of Multi-Protocol Label Switching (MPLS). We have done simulation in Matlab environment to compare the performance of MPLS against the IP network in a simulated environment. MPLS is a modern technique for forwarding network data. It broadens routing according to path controlling and packet forwarding. In this thesis MPLS is computed on the basis of its performance, efficiency for sending data from source to destination. A MATLAB based simulation tool is developed to compare MPLS with IP network in a simulated environment. The results show the performance of MPLS network in comparison of IP network. / YASSAR AZIZ, +46 73 9135296 MUHAMMAD NAEEM ASLAM +46 70 4918985

Thesis (MSc)--Stellenbosch University, 2003. / ENGLISH ABSTRACT: The Internet has experienced tremendous growth in the last three decades and has emerged as a platform to carryall forms of communications including voice, video and data. Along with this growth came the urgency for quality of service (QoS) controls in IP networks as different types of traffics have different service requirements. Although the IP protocol is able to scale to very large networks, it does not provide sufficient functionality for traffic engineering in order to enable QoS control. Multi-protocol label switching (MPLS) is a new routing technology that enhances IP with some QoS concepts from ATM and uses relatively simple packet forwarding mechanisms. MPLS has the ability to perform traffic engineering and QoS control by routing traffic flowson virtual connections called label switched paths (LSPs) which are assigned capacity. A large portion of the traffic carried on the Internet consists of data traffic in the form of TCP traffic. This thesis investigates several TCP models to find the ones most suitable to represent TCP traffic in MPLS networks. The models consist of three types. The first type models a single TCP source and the second type models a fixed number of TCP sources. The third type models an infinite number of TCP sources. The models were evaluated by comparing their throughput predictions and results obtained from simulation experiments that were done with the widely-used simulator ns. We also present a simple derivation of the 1/,;e law for the TCP congestion window size where e is the packet loss probability. / AFRIKAANSE OPSOMMING:In die afgelope drie dekades het die Internet beduidende groei ervaar, soveel so dat dit ontluik het as 'n medium om alle tipes van moderne kommunikasies te hanteer insluitend telefoon, video en data. Hierdie groei het gepaard gegaan met die behoefte na diensvlak (QoS) meganismes in IP netwerke aangesien verskillende tipe kommunikasies verskillende diens vereistes het. Alhoewel die IP protokol skalleerbaar is tot baie groot netwerke, voorsien dit nie voldoende funksionaliteit om QoS beheer toe te pas nie. "Multi-protocol label switching" (MPLS) is 'n nuwe roeterings tegnologie wat IP aanvul met QoS konsepte van ATM en dit maak gebruik van relatief eenvoudige pakkie versendings-meganismes. MPLS het die vermoe om netwerk-verkeer reeling en QoS beheer toe te pas deur verkeers-strome te roeteer op virtuele roetes genaamd "label switched paths" (LSPs) aan wie kapasiteit toegeken is. 'n Beduidende gedeelte van Internet-verkeer bestaan uit TCP-verkeer. Hierdie tesis ondersoek verskillende modelle van TCP om die te vind wat die mees geskik is om TCP verkeer in MPLS netwerke te verteenwoordig. Drie tipes modelle is ondersoek. Die eerste tipe moduleer 'n enkele TCP verkeersbron en die tweede tipe moduleer 'n vasgestelde aantal TCP verkeersbronne. Die derde tipe moduleer 'n oneindige aantal verkeersbronne. Die modelle is geevalueer deur hul voorspellings van die tempo van data transmissie te vergelyk met resultate van simulasies. Die simulasies is gedoen met die veelgebruikte simulator ns. Hierdie tesis bevat ook 'n eenvoudige afleiding vir die 1/,;e wet vir die TCP oorlading venster grootte met e die verlies waarskeinlikheid van 'n netwerk pakkie.

In response to the development of a next-generation networking (NGN) generalized multi-protocol label switching (GMPLS) technology is required for automatically switched optical network (ASON). Reconfigurable optical add-drop multiplexer (ROADM) is an indispensable device for the ASON, and the dense wavelength division multiplexed (DWDM) signals can be transmitted through the network under the management of the network administrator to configure dynamic customer needs and the desired quality of service (QoS). The ROADM can also increase the efficiency of utilizing the existing capacity of the optical fiber lines and can reduce or waive to set up additional optical fiber lines. This thesis studies a novel ROADM based on the arrayed waveguide grating (AWG) and the fiber Bragg grating (FBG) to overcome that the current ROADM cannot process that the input signal channels is greater than the wavelengths channels of AWG. Many types of ROADMs have been proposed and realized through different optical devices. Among these, hybrid optical circulator and FBG based ROADM is more attractive because of its low crosstalk and polarization insensitivity. However, it still suffers from many component counts and high insertion losses due to the use of many circulators and a multiplexer-demultiplexer pair. In this master thesis, we focus on demonstrating a novel ROADM and evaluating its crosstalk performance. It is found that the proposed ROADM has the advantage on extending the number of wavelength signal to make the system configurable and flexible.