Várias características da Internet mudaram drasticamente desde que o TCP foi criado, como o maior compartilhamento de recursos devido à maior quantidade de usuários, maior largura de banda disponível, a existência de muitas conexões que podem percorrer longas distâncias e a ubiquidade das redes sem fio. Confrontado com essas novas características, o TCP apresenta diversas limitações. Dentre elas estão a subutilização da rede quando a largura de banda é da ordem de centenas de Gbps, o favorecimento de conexões que possuem pouco atraso (poucas dezenas de milisegundos), a incapacidade de distinção de causas de perdas de pacote e a demora para estabelecimento de conexões seguras (até 3 RTTs). Nesse contexto, com o objetivo de tornar o transporte de dados na Internet mais rápido e eficiente, a Google desenvolveu o protocolo QUIC. O QUIC propõe diversos avanços em relação aos protocolos existentes, como um novo mecanismo para estabelecimento de conexão e controle de congestionamento otimizado. Resultados apresentados pela Google mostraram claro ganho de desempenho em relação ao TCP, justificando o trabalho de tornar o QUIC um padrão IETF da Internet. Porém, esses resultados são impossíveis de serem verificados porque nos relatórios divulgados não há informação suficiente para que os cenários de teste sejam reproduzidos e porque é implausível possuir a mesma infraestrutura para os testes que a Google tem. Neste trabalho, avaliamos o desempenho do protocolo QUIC em diversos cenários de rede, comparando-o com o desempenho de várias implementações do TCP, principalmente o CUBIC. Diferente do realizado na literatura, todos os cenários utilizados são bem descritos, permitindo a reprodutibilidade dos experimentos. Além disso, para a realização dos experimentos foi criado um novo módulo que implementa o QUIC no simulador de redes NS-3. Este módulo está disponível como software livre, permitindo que outros pesquisadores usem o módulo para replicar e verificar nossos experimentos e para criarem novos experimentos de forma reprodutível. Ademais, eles também podem usar o módulo como uma ferramenta para avaliar, de maneira rápida, o comportamento de novas técnicas dentro do protocolo. / Many characteristics of the Internet have drastically changed since TCP was created such as the increase on resource sharing due to a larger number of Internet users, the growth of available bandwidth, the existence of many connections that may travel long distances and the ubiquity of wireless networks. When faced with those new characteristics, TCP showed severe limitations. Among them are network underutilization in high bandwidth environments of hundreds of Gbps, favoring of connections with small delays (few tens of milliseconds), incapacity of distinguishing packet loss causes and high delays for establishing secure connections (up to 3 RTTs). In this context, with the goal of making Internet data transport faster and more efficient, Google has developed the QUIC protocol. QUIC proposes many advances compared to existing protocols, such as a new mechanism for establishing connections and an optimized congestion control algorithm. Google has reported results indicating that QUIC performs better than TCP, justifying the work on making QUIC an IETF Internet standard. However, those results cannot be verified because on the published reports there is not enough information to reproduce the test scenarios and it is implausible to have the same test infrastructure Google has. In this work, we evaluate QUICs performance in a number of network scenarios, comparing it with the performance of different TCP flavours, specially TCP CUBIC. Unlike other works in the literature, all scenarios are well described, enabling experiment replicability. Furthermore, to run experiments we created a new module that implements QUIC on the network simulator NS-3. The module is available as free software, allowing other researchers to use it to reproduce and verify our experiments and to create new ones in a replicable way. Additionally, they can use the module as a tool to quickly assess the behaviour of new techniques in the protocol.

Wireless ad-hoc networks are based on a cooperative communication model, where all nodes not only generate traffic but also help to route traffic from other nodes to its final destination. In such an environment where there is no infrastructure support the lifetime of the network is tightly coupled with the lifetime of individual nodes. Most of the devices that form such networks are battery-operated, and thus it becomes important to conserve energy so as to maximize the lifetime of a node. In this thesis, we present JTP, a new energy-aware transport protocol, whose goal is to reduce power consumption without compromising delivery requirements of applications. JTP has been implemented within the JAVeLEN system. JAVeLEN~\cite{javelen08redi}, is a new system architecture for ad hoc networks that has been developed to elevate energy efficiency as a first-class optimization metric at all protocol layers, from physical to transport. Thus, energy gains obtained in one layer would not be offset by incompatibilities and/or inefficiencies in other layers. To meet its goal of energy efficiency, JTP (1) contains mechanisms to balance end-to-end vs. local retransmissions; (2) minimizes acknowledgment traffic using receiver regulated rate-based flow control combined with selected acknowledgments and in-network caching of packets; and (3) aggressively seeks to avoid any congestion-based packet loss. Within this ultra low-power multi-hop wireless network system, simulations and experimental results demonstrate that our transport protocol meets its goal of preserving the energy efficiency of the underlying network. JTP has been implemented on the actual JAVeLEN nodes and its benefits have been demoed on a real system.

Thesis (M.S.)--Worcester Polytechnic Institute. / Keywords: TCP; RED; satellite; PEP; STP; performance enhancing proxy; segment caching; IP-ABR; Internet; bandwidth reservation; IP-VBR; congestion avoidance; bandwidth sharing. Includes bibliographical references (p. 98-99).

<p>The soaring demands for always-on and fast-response online services have driven modern datacenter networks to undergo tremendous growth. These networks often rely on scale-out designs with large numbers of commodity switches to reach immense capacity while keeping capital expenses under check. Today, datacenter network operators spend tremendous time and efforts on two key challenges: 1) how to efficiently utilize the bandwidth connecting host pairs and 2) how to promptly handle network failures with minimal disruptions to the hosted services.</p><p>To resolve the first challenge, we propose solutions in both network layer and transport layer. In the network layer solution, We advocate to design practical datacenter architectures for easy operation, i.e., an architecture should be reliable, capable of improving bisection bandwidth, scalable and debugging-friendly. By strictly following these four guidelines, We propose DARD, a Distributed Adaptive Routing architecture for Datacenter networks. DARD allows each end host to reallocate traffic from overloaded paths to underloaded paths without central coordination. We use congestion game theory to show that DARD converges to a Nash equilibrium in finite steps and its gap to the optimal flow allocation is bounded in the order of 1/logL, with L being the number of links. We use a testbed implementation and simulations to show that DARD can achieve a close-to-optimal flow allocation with small control overhead in practice.</p><p>In the transport layer solution, We propose Explicit Multipath Congestion Control Protocol (MPXCP), which achieves four desirable properties: fast convergence, efficiency, being fair to flows with different RTTs and negligible queue size. Intensive ns-2 simulation shows that MPXCP can quickly converge to efficiency and fairness without building up queues despite different delay-bandwidth products.</p><p>To resolve the second challenge, recent research efforts have focused on automatic failure localization. Yet, resolving failures still requires significant human interventions, resulting in prolonged failure recovery time. Unlike previous work, we propose NetPilot, a system aims to quickly mitigate rather than resolve failures. NetPilot mitigates failures in much the same way operators do -- by deactivating or restarting suspected offending components. NetPilot circumvents the need for knowing the exact root cause of a failure by taking an intelligent trial-and-error approach. The core of NetPilot is comprised of an Impact Estimator that helps guard against overly disruptive mitigation actions and a failure-specific mitigation planner that minimizes the number of trials. We demonstrate that NetPilot can effectively mitigate several types of critical failures commonly encountered in production datacenter networks.</p> / Dissertation

The Web has outgrown the transport mechanisms that have been used since its inception. Due to the increasing complexity of web pages in terms of both total size and number of individual resources, HTTP over TCP can no longer provide a satisfactory user performance. In recent years, much progress has been made in this area by evolving the web's underlying mechanisms. Multipath QUIC (MPQUIC) is one such approach. MPQUIC is a new transport protocol which enables multihomed devices, such as smartphones, to aggregate their network interfaces in order to achieve greater performance. Additionally, MPQUIC is capable of multiplexing several data streams concurrently over a single connection, which can also provide performance benefits. This work began with a validation of our MPQUIC setup, which was performed by comparing MPQUIC to another multipath solution in a large set of experiments. The results show that MPQUIC is generally beneficial for the transfer time of large files, which corresponds with results from previous works. We additionally investigated ways to exploit MPQUIC's multipath and stream features to achieve lower latencies for web pages via the means of packet scheduling. We implemented the Earliest Completion First (ECF) scheduler, and investigated how it compares against MPQUIC's default path scheduler. The results indicate that the ECF scheduler is significantly more capable of handling heterogeneous network scenarios than the default scheduler, and can achieve higher throughput and lower latencies. Next, a Stream Priority scheduler was designed and implemented, which utilizes stream priorities to achieve lower completion times for select streams. The results from the investigation indicate that proper stream scheduling can significantly reduce download times of the prioritized resources. This effect was especially noticeable as path characteristics diverge. We also show that proper configuration of stream priorities is critical for such a scheduler, as a sub-optimal configuration yielded poor performance.

The profileration of multimedia applications, such as streaming video, teleconferencing, and interactive gaming has created a tremendous challenge for the traditional transport protocols of the Internet – UDP and TCP. Specifically, many multimedia applications are examples of soft real-time applications. They have often relatively stringent require- ments in terms of delay and delay jitter, but typically tolerate a limited packet loss rate. In recognition of the transport service requirements of soft real-time applications, this thesis studies the feasibility of using retransmission based, partially reliable trans- port protocols for these applications. The thesis studies ways of designing retransmis- sion based, partially reliable transport protocols that are congestion aware and TCP com- patible. Furthermore, the transport protocols should provide a service that, in terms of performance metrics such as throughput, delay, and delay jitter, are suitable for soft real- time applications. The thesis work comprises the design, analysis, and evaluation of two retransmission based, partially reliable transport protocols: PRTP and PRTP-ECN. Extensive simulations have been carried out on PRTP as well as PRTP-ECN. These sim- ulations have in part been complemented by some theoretical analysis. The results of the simulations and the analysis suggest that substantial reductions in delay jitter and improvements in throughput can indeed be obtained with both PRTP and PRTP-ECN as compared to TCP. While PRTP reacted too slowly to congestion to be TCP-friendly and altogether fair, PRTP-ECN was found to be both TCP-friendly and reasonably fair. The thesis work also comprises an extensive survey on retransmission based, par- tially reliable transport protocols. Based on this survey, we have proposed a taxonomy for these protocols. The taxonomy considers two dimensions of retransmission based, partially reliable transport protocols: the transport service, and the error control scheme.

Les communications multimedia à forte contrainte de délai dominent de plus en plus l'Internet et supportent principalement des services interactifs et de diffusion de contenus multimedia. En raison de la prolifération des périphériques mobiles, les liaisons sans-fil prennent une part importante dans la transmission de ces données temps réel. Un certain nombre de questions doivent cependant être abordées afin d'obtenir une qualité de service acceptable pour ces communications dans ces environnements sans-fil. En outre, la disponibilité de multiples interfaces sans-fil sur les appareils mobiles offre une opportunité d'améliorer la communication tout en exacerbant encore les problèmes déjà présents sur les liaisons sans-fil simples. Pour faire face aux problèmes d'erreurs et de délai de ces liaisons, cette thèse propose deux améliorations possibles. Tout d'abord, une technique d'amélioration pour le protocole de transport multimedia Datagram Congestion Control Protocol (DCCP/CCID4) pour lien long délai (par exemple, lien satellite) qui améliore significativement les performances du transport de la voix sur IP (VoIP). En ce qui concerne les erreurs de lien et le multi-chemin, cette thèse propose et évalue un code à effacement adapté au protocole Stream Transport Control Protocol (CMT-SCTP). Enfin, un outil d'évaluation en ligne de streaming de qualité vidéo, implémentant une méthode cross-layer d'évaluation de la qualité vidéo en temps-réel pour encodeur H.264 est proposé à la communauté réseau en open-source. / Real time communications have, in the last decade, become a highly relevant component of Internet applications and services, with both interactive (voice and video) communications and streamed content being used by people in developed and developing countries alike. Due to the proliferation of mobile devices, wireless media is becoming the means of transmitting a large part of this increasingly important real time communications traffic. Wireless has also become an important technology in developing countries (and remote areas of developed countries), with satellite communications being increasingly deployed for traffic backhaul and ubiquitous connection to the Internet. A number of issues need to be addressed in order to have an acceptable service quality for real time communications in wireless environments. In addition to this, the availability of multiple wireless interfaces on mobile devices presents an opportunity to improve and further exacerbates the issues already present on single wireless links. Mitigation of link errors originating from the wireless media, addressing packet reordering, jitter, minimising the link and buffering (required to deal with reordering or jitter) delays, etc. all contribute to lowering user's quality of experience and perception of network quality and usability. Therefore in this thesis, we consider improvements to transport protocols for real time communications and streaming services to address these problems.

While developing networking for games and applications, developers have a list of network specific requirements to be met as well as decide how to meet them. It is not always easy to decide what protocol is best suited for a given network configuration, or what is the best size of a data packet. By performing a comparative analysis, it becomes possible to identify how protocols, packet size, and network configuration impact the one-way travel time and throughput of a given implementation. The result shows how the different implementations compared against each other and the analysis tries to determine why they perform as they do. This gives a good overview of the pros and cons of how TCP, TCP(N), UDP, and RakNet, behave and perform over LANs and WLANs with varying packet size.

In the last decade, we have seen an explosive growth in the deployment of multimedia applications on the Internet. However, the transport service provided over the Internet is not always feasible for these applications, since the network was originally designed for other types of applications. One way to better accommodate the service requirements of some of these applications is to provide a partially reliable transport service. A partially reliable transport service does not insist on recovering all, but just some of the packet losses, thus providing a lower transport delay than a reliable transport service. The work in this thesis focuses on the design, implementation, and evaluation of a partially reliable transport protocol called PRTP. PRTP has been designed as an extension to TCP in order to show that such a service could be effectively integrated with current protocol standards. An important feature of PRTP is that all modifications for PRTP are restricted to the receiver side, which means that it could be very easily deployed. The thesis presents performance results from various experiments on a Linux implementation of PRTP. The results suggest that transfer times can be decreased significantly when using PRTP as opposed to TCP in networks in which packet loss occurs. Furthermore, the thesis includes a study that investigates how users perceive an application that is based on a partially reliable service. Specifically, how users select the trade-off between image quality and latency when they download Web pages is explored. The results indicate that many of the users in the study could accept less than perfect image quality if the latency could be shortened.

Várias características da Internet mudaram drasticamente desde que o TCP foi criado, como o maior compartilhamento de recursos devido à maior quantidade de usuários, maior largura de banda disponível, a existência de muitas conexões que podem percorrer longas distâncias e a ubiquidade das redes sem fio. Confrontado com essas novas características, o TCP apresenta diversas limitações. Dentre elas estão a subutilização da rede quando a largura de banda é da ordem de centenas de Gbps, o favorecimento de conexões que possuem pouco atraso (poucas dezenas de milisegundos), a incapacidade de distinção de causas de perdas de pacote e a demora para estabelecimento de conexões seguras (até 3 RTTs). Nesse contexto, com o objetivo de tornar o transporte de dados na Internet mais rápido e eficiente, a Google desenvolveu o protocolo QUIC. O QUIC propõe diversos avanços em relação aos protocolos existentes, como um novo mecanismo para estabelecimento de conexão e controle de congestionamento otimizado. Resultados apresentados pela Google mostraram claro ganho de desempenho em relação ao TCP, justificando o trabalho de tornar o QUIC um padrão IETF da Internet. Porém, esses resultados são impossíveis de serem verificados porque nos relatórios divulgados não há informação suficiente para que os cenários de teste sejam reproduzidos e porque é implausível possuir a mesma infraestrutura para os testes que a Google tem. Neste trabalho, avaliamos o desempenho do protocolo QUIC em diversos cenários de rede, comparando-o com o desempenho de várias implementações do TCP, principalmente o CUBIC. Diferente do realizado na literatura, todos os cenários utilizados são bem descritos, permitindo a reprodutibilidade dos experimentos. Além disso, para a realização dos experimentos foi criado um novo módulo que implementa o QUIC no simulador de redes NS-3. Este módulo está disponível como software livre, permitindo que outros pesquisadores usem o módulo para replicar e verificar nossos experimentos e para criarem novos experimentos de forma reprodutível. Ademais, eles também podem usar o módulo como uma ferramenta para avaliar, de maneira rápida, o comportamento de novas técnicas dentro do protocolo. / Many characteristics of the Internet have drastically changed since TCP was created such as the increase on resource sharing due to a larger number of Internet users, the growth of available bandwidth, the existence of many connections that may travel long distances and the ubiquity of wireless networks. When faced with those new characteristics, TCP showed severe limitations. Among them are network underutilization in high bandwidth environments of hundreds of Gbps, favoring of connections with small delays (few tens of milliseconds), incapacity of distinguishing packet loss causes and high delays for establishing secure connections (up to 3 RTTs). In this context, with the goal of making Internet data transport faster and more efficient, Google has developed the QUIC protocol. QUIC proposes many advances compared to existing protocols, such as a new mechanism for establishing connections and an optimized congestion control algorithm. Google has reported results indicating that QUIC performs better than TCP, justifying the work on making QUIC an IETF Internet standard. However, those results cannot be verified because on the published reports there is not enough information to reproduce the test scenarios and it is implausible to have the same test infrastructure Google has. In this work, we evaluate QUICs performance in a number of network scenarios, comparing it with the performance of different TCP flavours, specially TCP CUBIC. Unlike other works in the literature, all scenarios are well described, enabling experiment replicability. Furthermore, to run experiments we created a new module that implements QUIC on the network simulator NS-3. The module is available as free software, allowing other researchers to use it to reproduce and verify our experiments and to create new ones in a replicable way. Additionally, they can use the module as a tool to quickly assess the behaviour of new techniques in the protocol.