Networks within data centers, such as connections between servers and disk arrays, need lossless flow control allowing all packets to move quickly through the network to reach their destination. This paper proposes a new algorithm for congestion control to satisfy the needs of such networks and to answer the question: Is it possible to provide circuit-less reliability and flow control in an Ethernet network? TCP uses an end-to-end congestion control algorithm, which is based on end-to-end round trip time (RTT). Therefore its flow control and error detection/correction approach is dependent on end-to-end RTT. Other approaches utilize specialized data link layer networks such as InfiniBand and Fibre Channel to provide network reliability. The algorithm proposed in this thesis builds on the ubiquitous Ethernet protocol to provide reliability at the data link layer without the overhead and cost of the specialized networks or the delay induced by TCP’s end-to-end approach. This approach requires modifications to the Ethernet switches to implement a back pressure based flow control algorithm. This back pressure algorithm utilizes a modified version of the Random Early Detection (RED) algorithm to detect congestion. Our simulation results show that the algorithm can quickly recover from congestion and that the average latency of the network is close to the average latency when no congestion is present. With correct threshold and alpha values, buffer sizes in the network and on the source nodes can be kept small to allow little needed additional hardware to implement the system.

Les ressources d'un réseau informatique sont partagées par un grand nombre d'utilisateurs, le réseau peut devenir congestionné voire saturé. Le contrôle de congestion est un élément critique garantissant son bon fonctionnement. Encore récemment, les problématiques d'ingénierie de trafic et de QoS n'étaient pas explorées dans le cadre des réseaux CCN (Content-Centric Networking).L'objectif de cette thèse consiste à évaluer l'impact de la congestion dans CCN, à concevoir un mécanisme de contrôle de congestion et à en évaluer l'efficacité.Nous avons réalisé une telle solution en deux temps. D'une part, nous proposons le mécanisme HoBHIS qui repose sur l'équilibrage de flot définie dans CCN. Ce mécanisme, déployé sur chaque routeur CCN, consiste à surveiller les échanges réalisés sur un même tampon de transmission. Il contrôle dynamiquement le taux d'envoi des demandes de contenu afin de garantir qu'au niveau du goulot d'étranglement, la taille de la file d'attente des paquets de donnée correspondante tende vers un seuil préconfiguré. D'autre part, nous proposons un mécanisme contrôlant le débit d'envoi d'un client afin d'éviter une congestion des files d'attente des noeuds CCN impliqués dans la communication.Nous avons prouvé mathématiquement l'efficacité de nos mécanismes et évalué leur performance à l'aide du simulateur ndnSIM.Les résultats de notre travail sont publiés dans des conférences internationales renommées. Nous avons aussi collaboré avec Cisco Systems (Boston, USA), qui s'est intéressé à nos travaux. Les résultats issus de cette collaboration sont publiés dans un article international qui a reçu un "Best Paper Award" et sont présentés dans cette thèse. / The network resources are shared between a large number of users. It may potentially create a risk for buffer overflow and performance degradation. That is why congestion control is critical to guarantee network performance. Congestion control schemes have been widely studied in the past but only recently in the context of CCN (Content-Centric Networking).This thesis explores the congestion control risk of CCN, identifies the bottlenecks and proposes strategies to circumvent them. We have designed our original hop-by-hop Interest shaping mechanism (HoBHIS) that nicely exploits the flow balance enforced in CCN between Interest and Chunk packets. It monitors active conversations sharing the transmission buffer of a CCN node face in order to dynamically adjust their Interest sending rate and enforce the Chunk queue length to converge to a defined objective. This mechanism is implemented in each CCN node. Then, we extended HoBHIS in order to address several important concerns that might occur in CCN. We proposed a Tolerance mechanism that controls the Clients sending rate as well as prevents the loss of Interest packets. We demonstrated the efficiency of our mechanisms through mathematical proofs and simulations performed in ndnSIM.Our results have been published in international conferences. The work has generated interest from the networking community. Particularly, "Cisco Systems" invited us to join their project on developing Named-Data Networking (NDN) traffic control mechanism. As one of the important results of this collaboration is a research article that got a SIGCOMM ICN workshop’13 “Best paper award”. This paper is also presented as a part of the dissertation.

行動隨意式網路(MANET)是一種具有高度動態拓撲結構的網路。每一個行動隨意式網路由一組移動節點(Node)組成，彼此之間互相支援轉送封包可以不依靠基地台建構成Intranet。此種網路中，因節點移動之緣故，連線不穩定、頻寬較窄，錯誤率亦較高。傳統的TCP傳輸協定在行動隨意式網路上的效能不免遭受重創。 傳統的TCP在封包遺失時，只能從傳送端進行重傳，而行動隨意式網路傳輸品質極不穩定，常常重送多次才可到達目的地，導致要耗費極長的時間才能將封包送達目的地，然而，在行動隨意式網路中，大量傳輸資料的需求並不大，反而是封包的快速送達更為重要，因此加速封包的送達成為比增大傳送量更為重要的目標。 為了使封包較快送達目的地端，我們提出了Hop-by-Hop TCP的方法，使每個節點使用當地重傳以保證封包成功的傳到下一個節點，遺失的封包不必重新由傳送端重傳，能更快反應封包遺失，並且提昇傳輸可靠度，使封包在高遺失率的情形之下能順利且較為快速的送達目的地端。 我們利用NS-2網路模擬器進行實驗，驗證我們的機制，實驗在不同的拓樸及負載等參數下進行，觀察傳輸成功率及封包傳輸時間，以及公平性。實驗結果指出，本方法在網路環境不穩定時吞吐量能有25.7%以上的提昇，而延遲時間也能有25%的提昇，亦有相當好的公平性。 / A Mobile Ad hoc Network (MANET) MANET is composed of a group of mobile computing devices (nodes) that are equipped with Wireless LAN (WLAN) capability. Nodes can transmit packets to each other to construct Intranet without any base station. In an MANET environment, the communication links are unstable due to various reasons. Error rate is higher and bandwidth is smaller than fixed networks. Running regular TCP protocol on MANET will suffer from serious performance degradation in MANET. To handle packet lost, regular TCP can only retransmit lost packets from the source. However, when error rate is high, several retransmissions may be needed to transmit a packet to its destination successfully. As a result, the effective bandwidth is much lower and the average time to transmit a packet will be much longer. Considering that most applications on MANET prefer shorter transmission time to higher bandwidth, this thesis proposes Hop-by-Hop TCP protocol aiming to accelerate the transmission of packets. Hop-by-Hop TCP makes every intermediate node in the transmission path running a local TCP to guarantee the transmission of each packet on each link. The retransmission of a lost packet is right at the transmitting end of the link where the packet is lost. It doesn't need to retransmit a lost packet from its source node. It takes less time in average to transmit a packet to its destination in a high error rate environment. We evaluate the performance of our approach by simulation using NS-2 simulator. Our experiments show that our proposed protocol outperforms TCP Reno by 25.7% in throughput and 25% reduction in average transmission time. The fairness requirement is also achieved while our proposed protocol coexists with other major TCP variants.