Hoje em dia, distribuições de grandes volumes de dados por redes TCP/IP corporativas trazem problemas como a alta utilização da rede e de servidores, longos períodos para conclusão e maior sensibilidade a falhas na infraestrutura de rede. Estes problemas podem ser reduzidos com utilização de redes par-a-par (P2P). O objetivo desta dissertação é analisar o desempenho do protocolo BitTorrent padrão em redes corporativas e também realizar a análise após uma modificação no comportamento padrão do protocolo BitTorrent. Nesta modificação, o rastreador identifica o endereço IP do par que está solicitando a lista de endereços IP do enxame e envia somente aqueles pertencentes à mesma rede local e ao semeador original, com o objetivo de reduzir o tráfego em redes de longa distância. Em cenários corporativos típicos, as simulações mostraram que a alteração é capaz de reduzir o consumo médio de banda e o tempo médio dos downloads, quando comparados ao BitTorrent padrão, além de conferir maior robustez à distribuição em casos de falhas em enlaces de longa distância. As simulações mostraram também que em ambientes mais complexos, com muitos clientes, e onde a restrição de banda em enlaces de longa distância provoca congestionamento e descartes, o desempenho do protocolo BitTorrent padrão pode ser semelhante a uma distribuição em arquitetura cliente-servidor. Neste último caso, a modificação proposta mostrou resultados consistentes de melhoria do desempenho da distribuição. / Nowadays, distributions of large volumes of data over enterprise TCP/IP networks bring problems such as high network and server utilizations, long periods for completion, and greater sensitivity to flaws in network infrastructure. These problems can be reduced with the use of Peer-to-Peer networks (P2P). The aim of this work is to analyze the performance of the standard BitTorrent protocol in corporate networks and also perform the analysis after a change in the default behavior of the BitTorrent protocol. In this modification, the tracker identifies the peer IP address requesting the list of IP addresses of the swarm and sends only those belonging to the same LAN and to the original seeder, with the aim of reducing traffic on WAN links. In typical enterprise scenarios, the simulations showed that the change is able to reduce the average bandwidth consumption and the average time of downloads compared with standard BitTorrent, and give greater robustness to the distribution in case of failure of WAN links. The simulations also showed that in more complex network environments, with many clients, and where the bandwidth restriction on long distance links causes congestion and packet drops, the performance of standard BitTorrent protocol can be similar than a distribution in client-server architecture. In the latter case, the proposed change showed consistent results in improving distribution performance.

Hoje em dia, distribuições de grandes volumes de dados por redes TCP/IP corporativas trazem problemas como a alta utilização da rede e de servidores, longos períodos para conclusão e maior sensibilidade a falhas na infraestrutura de rede. Estes problemas podem ser reduzidos com utilização de redes par-a-par (P2P). O objetivo desta dissertação é analisar o desempenho do protocolo BitTorrent padrão em redes corporativas e também realizar a análise após uma modificação no comportamento padrão do protocolo BitTorrent. Nesta modificação, o rastreador identifica o endereço IP do par que está solicitando a lista de endereços IP do enxame e envia somente aqueles pertencentes à mesma rede local e ao semeador original, com o objetivo de reduzir o tráfego em redes de longa distância. Em cenários corporativos típicos, as simulações mostraram que a alteração é capaz de reduzir o consumo médio de banda e o tempo médio dos downloads, quando comparados ao BitTorrent padrão, além de conferir maior robustez à distribuição em casos de falhas em enlaces de longa distância. As simulações mostraram também que em ambientes mais complexos, com muitos clientes, e onde a restrição de banda em enlaces de longa distância provoca congestionamento e descartes, o desempenho do protocolo BitTorrent padrão pode ser semelhante a uma distribuição em arquitetura cliente-servidor. Neste último caso, a modificação proposta mostrou resultados consistentes de melhoria do desempenho da distribuição. / Nowadays, distributions of large volumes of data over enterprise TCP/IP networks bring problems such as high network and server utilizations, long periods for completion, and greater sensitivity to flaws in network infrastructure. These problems can be reduced with the use of Peer-to-Peer networks (P2P). The aim of this work is to analyze the performance of the standard BitTorrent protocol in corporate networks and also perform the analysis after a change in the default behavior of the BitTorrent protocol. In this modification, the tracker identifies the peer IP address requesting the list of IP addresses of the swarm and sends only those belonging to the same LAN and to the original seeder, with the aim of reducing traffic on WAN links. In typical enterprise scenarios, the simulations showed that the change is able to reduce the average bandwidth consumption and the average time of downloads compared with standard BitTorrent, and give greater robustness to the distribution in case of failure of WAN links. The simulations also showed that in more complex network environments, with many clients, and where the bandwidth restriction on long distance links causes congestion and packet drops, the performance of standard BitTorrent protocol can be similar than a distribution in client-server architecture. In the latter case, the proposed change showed consistent results in improving distribution performance.

Observou-se nos últimos anos um grande aumetno na complexidade das redes. Surgiram também novos desa os para gerenciamento dessas redes. A dimensão atual e as tendências de crescimento das infraestruturas tem inviabilizado as técnicas de gerencimento de redes atuais, baseadas na intervenção humana. Por exemplo, a heterogeneidade dos elementos gerenciados obrigam que administradores e gerentes lidem com especi cidades de implanta ção que vão além dos objetivos gerenciais. Considerando as áreas funcionais da gerência de redes, a gerência de falhas apresenta impactos operacionais interessantes. Estima-se que 33% dos custos operacionais estão relacionados com a prevenção e recuperação de falhas e que aproximadamente 44% desse custo visa à resolução de problemas causados por erros humanos. Dentre as abordagens de gerência de falhas, o Self-Healing objetiva minimizar as interações humanas nas rotinas de gerenciamento de falhas, diminuindo dessa forma erros e demandas operacionais. Algumas propostas sugerem que o Self-Healing seja planejado no momento do projeto das aplicações. Tais propostas são inviáveis de aplicação em sistemas legados. Otras pesquisas sugerem à análise e instrumentação das aplicações em tempo de execução. Embora aplicáveis a sistemas legados, análise e instrumentação em tempo de execução estão fortemente acopladas as tecnologias e detalhes de implementação das aplicações. Por esse motivo, é difícil aplicar tais propostas, por exemplo, em um ambiente de rede que abrange muitas entidades gerenciadas implantadas através de diferentes tecnologias. Porém, parece plausível oferecer aos adminitradores e gerentes facilidades através das quais eles possam expressar seus conhecimentos sobre anoamlias e falhas de aplicações, bem como mecanismos através dos quais esses conhecimentos possam ser utilizado no gerenciamento de sistemas. Essa dissertação de mestrado tem como objetivo apresentar e avaliar uma solução comum que introduza nas redes capacidades de self-healing. A solu- ção apresentada utiliza-se de workplans para capturar o conhecimento dos administradores em como diagnosticar e recuperar anomalias e falhas em redes. Além disso, o projeto e implementação de um framework padrão para detecção e noti cação de falhas é discutido no âmbito de um sistema de gerenciamento baseado em P2P. Por último, uma avaliação experimental clari ca a viabilidade do ponto de vista operacional. / In recent years, a huge raise in networks' complexity was witnessed. Along the raise in complexity, many management challenges also arose. For instance, managed entities' heterogeneity demands administrators and managers to deal with cumbersome implementation and deployment speci cities. Moreover, infrastructures' current size and growth-trends show that it is becoming infeasible to rely on human-in-the-loop management techniques. Inside the problem domain of network management, Fault Management is appealing because of its impact in operational costs. Researches estimate that more than 33% of operational costs are related to preventing and recovering faults, where about 40% of this investment is directed to solve human-caused operational errors. Hence, addressing human interaction is mostly unarguably a need. Among di erent approaches, Self-Healing, a property of Autonomic Network Management's proposal, targets to avoid humans' interactions and decisions on Fault Management loops, thereupon unburden administrators and managers from performing Fault Management-related tasks. Some researches on Self-Healing enabling approaches suppose that Fault Management capabilities should be planned in design-time. These approaches are impossible to apply on legacy systems. Other researches suggest runtime analysis and instrumentation of applications' bytecode. Albeit applicable to some legacy systems, these last proposals are tightly-coupled to implementation's issues of underlaying technologies. For this reason, it is hard to apply such proposals end-toend, for example, in a scenario encompassing many managed entities implemented through di erent technologies. However, it is possible to o er to administrators and managers facilities to express they knowledge about networks' anomalies and faults, and facilities to leverage this knowledge. This master dissertation has as objective to present and evaluate a solution to imbue network management systems with self-healing capabilities. The solution relies on workplans as a mean to gather administrators and managers' knowledge on how to diagnose and heal networks' anomalies and faults. Besides that, the design and implementation of a standard framework for fault detection and noti cation customization is discussed while considering a P2P-Based Network Management System as its foundations. At last, an experimental evaluation renders clear the proposal's feasibility from the operational point of view.

Peer-to-Peer (P2P) overlays were initially proposed for use with wired networks. However, the very rapid proliferation of wireless communication technology has prompted a need for adoption of P2P systems in mobile networks too. There are many common characteristics between P2P overlay networks and Mobile Ad-hoc Networks (MANET). Self-organization, decentralization, a dynamic nature and changing topology are the most commonly shared features. Furthermore, when used together, the two approaches complement each other. P2P overlays provide data storage/retrieval functionality and MANET provides wireless connectivity between clients without depending on any pre-existing infrastructure. P2P overlay networks can be deployed over MANET to address content discovery issues. However, previous research has shown that deploying P2P systems straight over MANET does not exhibit satisfactory performance. Bandwidth limitation, limited resources and node mobility are some of the key constraints. This thesis proposes a novel approach, OneHopOverlay4MANET, to exploit the synergies between MANET and P2P overlays through cross-layering. It combines Distributed Hash Table (DHT) based structured P2P overlays with MANET underlay routing protocols to achieve one logical hop between any pair of overlay nodes. OneHopOverlay4MANET constructs a cross-layer channel to permit direct exchange of routing information between the Application layer, where the overlay operates, and the MANET underlay layer. Consequently, underlay routing information can be shared and used by the overlay. Thus, OneHopOverlay4MANET reduces the typical management traffic when deploying traditional P2P systems over MANET. Moreover, as a result of building one hop overlay, OneHopOverlay4MANET can eliminate the mismatching issue between overlay and underlay and hence resolve key lookups in a short time, enhancing the performance of the overlay. v In this thesis, we present OneHopOverlay4MANET and evaluate its performance when combined with different underlay routing protocols. OneHopOverlay4MANET has been combined with two proactive underlays (OLSR and BATMAN) and with three reactive underlay routing protocols (DSR, AODV and DYMO). In addition, the performance of the proposed system over OLSR has been compared to two recent structured P2P over MANET systems (MA-SP2P and E-SP2P) that adopted OLSR as the routing protocol. The results show that better performance can be achieved using OneHopOverlay4MANET.

Observou-se nos últimos anos um grande aumetno na complexidade das redes. Surgiram também novos desa os para gerenciamento dessas redes. A dimensão atual e as tendências de crescimento das infraestruturas tem inviabilizado as técnicas de gerencimento de redes atuais, baseadas na intervenção humana. Por exemplo, a heterogeneidade dos elementos gerenciados obrigam que administradores e gerentes lidem com especi cidades de implanta ção que vão além dos objetivos gerenciais. Considerando as áreas funcionais da gerência de redes, a gerência de falhas apresenta impactos operacionais interessantes. Estima-se que 33% dos custos operacionais estão relacionados com a prevenção e recuperação de falhas e que aproximadamente 44% desse custo visa à resolução de problemas causados por erros humanos. Dentre as abordagens de gerência de falhas, o Self-Healing objetiva minimizar as interações humanas nas rotinas de gerenciamento de falhas, diminuindo dessa forma erros e demandas operacionais. Algumas propostas sugerem que o Self-Healing seja planejado no momento do projeto das aplicações. Tais propostas são inviáveis de aplicação em sistemas legados. Otras pesquisas sugerem à análise e instrumentação das aplicações em tempo de execução. Embora aplicáveis a sistemas legados, análise e instrumentação em tempo de execução estão fortemente acopladas as tecnologias e detalhes de implementação das aplicações. Por esse motivo, é difícil aplicar tais propostas, por exemplo, em um ambiente de rede que abrange muitas entidades gerenciadas implantadas através de diferentes tecnologias. Porém, parece plausível oferecer aos adminitradores e gerentes facilidades através das quais eles possam expressar seus conhecimentos sobre anoamlias e falhas de aplicações, bem como mecanismos através dos quais esses conhecimentos possam ser utilizado no gerenciamento de sistemas. Essa dissertação de mestrado tem como objetivo apresentar e avaliar uma solução comum que introduza nas redes capacidades de self-healing. A solu- ção apresentada utiliza-se de workplans para capturar o conhecimento dos administradores em como diagnosticar e recuperar anomalias e falhas em redes. Além disso, o projeto e implementação de um framework padrão para detecção e noti cação de falhas é discutido no âmbito de um sistema de gerenciamento baseado em P2P. Por último, uma avaliação experimental clari ca a viabilidade do ponto de vista operacional. / In recent years, a huge raise in networks' complexity was witnessed. Along the raise in complexity, many management challenges also arose. For instance, managed entities' heterogeneity demands administrators and managers to deal with cumbersome implementation and deployment speci cities. Moreover, infrastructures' current size and growth-trends show that it is becoming infeasible to rely on human-in-the-loop management techniques. Inside the problem domain of network management, Fault Management is appealing because of its impact in operational costs. Researches estimate that more than 33% of operational costs are related to preventing and recovering faults, where about 40% of this investment is directed to solve human-caused operational errors. Hence, addressing human interaction is mostly unarguably a need. Among di erent approaches, Self-Healing, a property of Autonomic Network Management's proposal, targets to avoid humans' interactions and decisions on Fault Management loops, thereupon unburden administrators and managers from performing Fault Management-related tasks. Some researches on Self-Healing enabling approaches suppose that Fault Management capabilities should be planned in design-time. These approaches are impossible to apply on legacy systems. Other researches suggest runtime analysis and instrumentation of applications' bytecode. Albeit applicable to some legacy systems, these last proposals are tightly-coupled to implementation's issues of underlaying technologies. For this reason, it is hard to apply such proposals end-toend, for example, in a scenario encompassing many managed entities implemented through di erent technologies. However, it is possible to o er to administrators and managers facilities to express they knowledge about networks' anomalies and faults, and facilities to leverage this knowledge. This master dissertation has as objective to present and evaluate a solution to imbue network management systems with self-healing capabilities. The solution relies on workplans as a mean to gather administrators and managers' knowledge on how to diagnose and heal networks' anomalies and faults. Besides that, the design and implementation of a standard framework for fault detection and noti cation customization is discussed while considering a P2P-Based Network Management System as its foundations. At last, an experimental evaluation renders clear the proposal's feasibility from the operational point of view.

Made available in DSpace on 2015-04-11T14:03:12Z (GMT). No. of bitstreams: 1 DISSERTACAO ANDERSON.pdf: 945892 bytes, checksum: 13a676735f1c713ef1251e829d5e41bb (MD5) Previous issue date: 2006-09-13 / Fundação de Amparo à Pesquisa do Estado do Amazonas / More and more the demanding for mobility has been wanted by the users who want to have everything no matter where they are. The applications have been modified in order to attend this needs demanded by the new environment without any predictable infrastructure or topology. The adhoc networks are very popular when concerning about non-infrastructures environment and many studies have been started looking for solutions and strategies to understand and adapt the applications originally created to work over client/server architecture. Voice over IP applications also have to be adapted to work properly over this environment and this study try to identify the factors which can impact on application performance. It is made looking at the look up algorithms used by them to find other nodes around the net. / A medida que a mobilidade se torna uma exigência cada vez maior dos usuários de redes de computadores, as aplicações precisam adaptarem-se ao novo ambiente descentralizado, que por vezes, é totalmente imprevisível e mutável. As redes adhoc estão muito difundidas em cenários totalmente descentralizados e desprovidos de infra-estrutura fixa e por isso diversos estudos estão sendo realizados visando a adaptação de aplicações originalmente construídas sobre a luz da arquitetura cliente/servidor. As aplicações de voz sobre IP (VoIP) também precisam se adaptar a nova realidade, e para isso foram adotadas diversas estratégias de descentralização, sendo uma delas a utilização de algoritmos de busca P2P para prover a independência de pontos centralizadores. Este trabalho realiza um estudo do comportamento desse tipo de algoritmo sobre redes adhoc visando identificar os fatores que podem influênciar direta ou indiretamente a performance de uma aplicação VoIP que os utilize como m´método de localização de seus nós.

Submitted by Renata Lopes (renatasil82@gmail.com) on 2017-05-30T20:25:10Z No. of bitstreams: 1 franciscohenriquecerdeiraferreira.pdf: 1810078 bytes, checksum: b5d9796bdd636be5e4f0f3212220ad0a (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2017-06-01T11:36:20Z (GMT) No. of bitstreams: 1 franciscohenriquecerdeiraferreira.pdf: 1810078 bytes, checksum: b5d9796bdd636be5e4f0f3212220ad0a (MD5) / Made available in DSpace on 2017-06-01T11:36:20Z (GMT). No. of bitstreams: 1 franciscohenriquecerdeiraferreira.pdf: 1810078 bytes, checksum: b5d9796bdd636be5e4f0f3212220ad0a (MD5) Previous issue date: 2013-03-11 / Nos últimos anos, as aplicações P2P de transmissão de vídeo ao vivo despertaram um grande interesse na comunidade cientí ca. Essas aplicações geram um grande volume de dados que afetam diretamente o desempenho da rede. Apesar de existir um grande número de trabalhos dedicados a entender as aplicações P2P de transmissão de vídeo ao vivo, a maioria deles se baseia em uma visão estática desses sistemas. Estes trabalhos não se preocupam em entender a dinâmica do sistema, ou seja, como eles evoluem ao longo do tempo. Através de experimentos realizados na plataforma PlanetLab, este trabalho apresenta uma caracterização da aplicação SopCast, um dos mais importantes sistemas P2P de transmissão de vídeo ao vivo. O estudo é realizado levando-se em consideração a dinamicidade da topologia construída. Também é investigada a formação de comunidades na rede sobreposta e a correlação dessas comunidades com os Sistemas Autônomos da Internet (AS). Os resultados mostram que a formação de comunidades é bem de nida pela troca de tráfego e uma pequena porção de participantes desses grupos é responsável por sustentar toda a comunidade. Além disso, não existe indícios que os participantes se agrupam de acordo com seus Sistemas Autônomos. De fato, a probabilidade de uma comunidade ser formada com mais da metade de membros pertencentes a um mesmo AS é inferior a 10%. Finalmente, as caracterizações apresentadas fornecem informações importantes para o desenvolvimento de novas aplicações P2P de transmissão de vídeo ao vivo. Mais que isso, será possível algoritimos de formação de parcerias e grupos de tal forma que seja possível reduzir os custos de transmissão entre redes diferentes. / P2P live streaming systems have attracted a lot of attention from the research community in the last years. Such systems generate a large amount of data which impacts the network performance. Despite large number of works devoted to understand P2P live streaming applications, most of them rely on characterizing the static view of these systems. These works do not worry about either understanding the systems dynamics or analyzing how it evolves over time. Through experiments in PlanetLab platform, we present a characterization of the SopCast application, one of the most important P2P live streaming system. We focus on characterizing the dynamics of P2P overlay topology. We also investigated the community formation process in SopCast overlay and the correlation of these communities with the Autonomous Systems of the Internet (AS). Our results show that the formation of communities is well de ned by the exchange of tra c and a small portion of peers of these groups is responsible for supporting the entire community. Furthermore, there is a evidence that peeers are not grouped according to their Autonomous Systems. Indeed, the probability of a community be formed with more 50% of members belonging to the same AS is less than 10%. Finally, the characterizations we conduct provide important information to new P2P live streaming protocols and membership algorithms design. Moreover, the presented characterization may be useful to developers create algorithms that reduces the transmission cost on the P2P network.

Les solutions de streaming vidéo adaptatives basées sur l’utilisation du protocol HTTP ont été largement plébiscitées dans les mondes de l’industrie et de la recherche, notamment pour les possibilités d’améliorations de qualité d’experience qu’elles offrent ainsi que pour leurs facilités de déploiement liées au protocol HTTP. Pour autant, bien que ces solutions permettent d’augmenter la qualité d’experience utilisateurs en diminuant la qualité de la vidéo transmise sur les réseaux pour minimiser les interruptions vidéo liées au temps de chargement, la qualité intrinsèque de la vidéo est limitée par les capacités physiques du chemin entre le serveur utilisé et le client. Dans l’objectif d’augmenter la qualité d’experience utilisateurs et de diminuer les couts de déploiements des services de streaming, les travaux de cette thèse de doctorat proposent de faire évoluer de façon pragmatique les solutions de streaming adaptatives actuelles vers l’utilisation en simultané de plusieurs sources (serveurs ou pairs). La première contribution de cette thèse présente MS-Stream, une technique évolutive de streaming adaptatif basé sur HTTP et utilisant plusieurs serveurs simultanément. MS-Stream offre la possibilité d’exploiter la bande passante disponible dans les infrastructures distribuées et les réseaux hétérogènes. La deuxieme contribution de ce document est MATHIAS, un groupe d’algorithmes d’adaptation centrés client, implémentés dans MS-Stream, qui a pour vocation d’optimiser l’utilisation des ressources réseau hétérogènes mises à disposition du client pour obtenir une qualité vidéo cible. MATHIAS permet à chaque client de controller le nombre de serveur utilisé en simultané, de faire face à l’hétérogeneité des resources disponibles, de réagir aux fluctuations soudaines et non-anticipées des capacités des serveurs tout en donnant à l’utilisateur une experience de streaming ininterrompu. Pour finir, nous allons plus loin dans les capacités de scalabilité et de qualité d’experience de MS-Stream et MATHIAS en tirant profit des ressources physiques des consommateurs. Nous proposons une solution hybride pair-à-pair/multi-server de streaming adaptative: PMS. Au sein de PMS, les logiques d’adaptation de la qualité vidéo et de la scalabilité sont distribuées pour permettre à chaque client de tendre vers une utilisation optimale de l’infrastructure de streaming. / Single-source HTTP Adaptive Streaming solutions (HAS) have become the de-facto solutions to deliver video over the Internet mostly due to their capabilities to increase end-user’s Quality of Experience (QoE) as well as their ease of deployment due to the usage of the HTTP protocol. Although HAS solutions can increase QoE by trading off the delivered video quality to minimize the number of video freezing events, they are limited by the bandwidth available on the considered communication channel between the client and the server. This thesis exposes our contributions in building lightweight pragmatic and evolving solutions advocating for the simultaneous usage of multiple sources with heterogeneous capacities so as to achieve high QoE content delivery at low cost. The first contribution of this work presents a streaming solution extending HAS capabilities to a pragmatic multi-server technique: MS-Stream. MS-Stream provides the means to exploit expanded bandwidth and link diversity in distributed heterogeneous network infrastructures. In our second contribution, we propose MATHIAS, a client-side two-phase consumption and adaptation algorithm implemented into MSStream. MATHIAS aims at increasing the end-user’s perceived streaming quality while utilizing the most of the heterogeneous capacities offered at the service and network environments. Finally, we further extend the QoE and scalability capabilities of MS-Stream and MATHIAS by leveraging on clients’ connectivity capacities and we expose our third contribution: a hybrid P2P/Multi-server live-Streaming system (PMS) incorporating distributed quality and scalability adaptation mechanisms.

Taking the representative P2P lending platform Renrendai as an example, this paper focuses on the impact of borrower's region on the behavior of lenders and borrowers in the market. According to the Chinese six geographical regions the borrowers are from, this paper empirically analyzes the difference of success rate of borrowing and default rate in the six regions using the binary logistic regression model and further studies the reasons behind the regional difference. The result shows that the impact of regional difference is significant and the borrower from northern China are more likely to fund successfully, but the impact of regional difference on the default rate is insignificant, and the economic, financial and education development level in regions have a significant impact on the success rate of borrowing. This paper studies the regulatory differences of P2P platforms in various regions of China, the result shows that eastern China, central and southern China, and Beijing (in northern China) have paid more attention and importance to the regulation of P2P platforms. Keywords ： China; P2P lending; the success rate of borrowing; regional difference; regulation policy

Peer-to-Peer(P2P)架構讓社群內的使用者收集分散在網路各處之資源，其中最為風行的P2P系統當屬P2P檔案分享系統。P2P檔案分享系統之架構可分為集中式及分散式，而分散式架構又可細分為結構化及非結構化兩種。採用分散式且非結構化之BitTorrent-like架構，因其可擴張性較佳而廣為風行。在經驗中發現，BitTorrent-like架構在非對稱網路之下雖擁有寬裕的下行頻寬，但是其頻寬使用率卻不高，我們分析其成因如下：(1) Fractional Upward Bandwidth (FUB) 、 (2)Blockage of Acknowledgement (BoA) 與 (3)Long Physical Paths 等。而其中的Blockage of ACK問題，現今尚未有研究學者提出完整解決方案。本研究之目的，即要針對此Blockage of ACK的問題，改良網路協定中的傳輸層 (Transport Layer) 協定以提昇P2P檔案分享系統之效能。我們改用UDP作為傳輸層協定，並在應用層加入自動重建遺失之封包、決定基本傳輸單位大小及決定資料傳送速率等機制，以彌補UDP之缺陷。文中並提供了傳輸協定運作時所需參數的估計法，並且也與其它傳輸協定做效能之比較。實驗結果發現，我們設計的傳輸協定確可改善P2P檔案共享系統的運作效能。 / Peer-to-Peer (P2P) architectures let participants gather resources form network and make participants acquire more computation resources than they could offer. One of the most prominent P2P systems is P2P file sharing system. P2P file sharing system could be classified into to 2 categories: centralized and decentralized model. BitTorrent-like (BT-like) model can be classified as decentralized and unstructured model. BT-like model is quit popular nowadays due to its scalability. Unfortunately, BT-like model has several shortcomings on performance over asymmetric networks, because of some problems, such as Fractional Upward Bandwidth (FUB), Blockage of Acknowledgement (BoA) and long physical paths. No complete solution is available yet to solve the BoA problem. We propose a new UDP-based protocol to alleviate this problem. Since UDP protocol is not able to guarantee the data-integrity nor to determine the proper transmission rate by itself. To offer a complete solution, we have to enhance UDP, to add extra capabilities such as packet loss recovery, segment size determination and data rate determination mechanisms. Experiments show that our proposed protocol has good improvement on performance.