Catálogo de publicaciones - libros

This paper presents a sender side TCP congestion control scheme that reduces biases in wired as well as wireless networks. TCP has a problem utilizing the full bandwidth in high speed networks with a long delay. Moreover, competing flows with different roundtrip times share the bandwidth unfairly; a flow with long RTT experiences a throughput penalty. The throughput penalty is severe in wireless networks since TCP treats packet losses caused by link error as an indication of network congestions that trigger transfer rate reductions. The proposed scheme controls the network congestion in two phases – a fair convergence phase and a congestion avoidance phase – both of which are based on the application’s transfer data patterns. The transfer rate is then adjusted adaptively by considering the current transfer rate and the estimated bandwidth in order to reduce bias and throughputs. The scheme has been implemented in the Linux platform and experimented with various TCP variants in real environments. The experimental results show that the mechanism reduces biases, and the network bandwidth is shared fairly among the proposed and the traditional TCP flows.

In this paper, we propose a novel congestion control mechanism of TCP, by using an inline network measurement technique. By using information of available bandwidth of a network path between sender and receiver hosts, we construct quite a different congestion control mechanism from the traditional TCP Reno and its variants, based on logistic and Lotka-Volterra models from biophysics. The proposed mechanism is intensively investigated through analysis and simulation evaluations, and we show the effectiveness of the proposed mechanism in terms of scalability with the network bandwidth, convergence time, fairness among connections, and stability.

Transmission Control Protocol (TCP) is a reliable transport protocol tuned to perform well in habitual networks made up of links with low bit-error rates. TCP was originally designed for wired networks, where packet loss is assumed to be due to congestion. In wireless links, packet losses are due to high error rates and the disconnections induced are due to mobility. TCP responds to these packet losses in the same way as wired links. It reduces the window size before packet retransmission, initiates congestion control avoidance mechanism and resets its transmission timer. This adjustment results in unnecessary reduction of the bandwidth utilization causing significant degraded end-to-end performance. A number of approaches have been proposed to improve the efficiency of TCP in an unreliable wireless network. But researches only focus on scenarios where TCP sender is a fixed host. In this paper we propose a novel protocol called V-TCP (versatile TCP), an approach that mitigates the degrading effect of host mobility on TCP performance. In addition to scenarios where TCP sender is fixed host, we also analyze the scenario where TCP sender is a mobile host. V-TCP modifies the congestion control mechanism of TCP by simply using the network layer feedback in terms of disconnection and connection signals, thereby enhancing the throughput in wireless mobile environments. Several experiments were performed using NS-2 simulator and the results were compared to the performance of V-TCP with Freeze-TCP [1], TCP Reno and with 3-dupacks [2]. Performance results show an improvement of up to 50% over TCP Reno in WLAN environments and up to 150% in WWAN environments in both directions of data transfer.

We study the unfairness problem for TCP Vegas. There are three sources to cause the unfairness for Vegas: < , over-estimation of base RTT, and multiple congested gateways. To solve the unfairness caused by multiple congested gateways, we propose a new version of Vegas–Adaptive Vegas(A-Vegas) which assigns the value of parameters adaptively according to the number of congested gateways. Simulation shows that A-Vegas not only can solve the unfairness caused by multiple congested gateways, it also reduces the unfairness caused by over-estimation. We also introduce a new fairness index, RNBS (normal bandwidth sharing ratio), which indicates the ratio of the amount of bandwidth grabbed by a connection to the uniformly distributed bandwidth.

This paper begins with a brief literature review of various approaches to congestion avoidance and control of Internet traffic. It focuses mainly on one of Active Queue Management (AQM) schemes known as Random Early Detection (RED) mechanism for congestion control at routers. Towards the end of paper, an approximate analytical performance model has been proposed based on standard RED mechanism as an effective congestion control technique. Methodology adopted is based on Principle of Maximum Entropy (ME) to model RED mechanism. To model the bursty input traffic, Generalized Exponential (GE) distribution has been used. Closed form expressions for the state and blocking probabilities have also been presented. Numerical examples have been presented. By comparing results obtained from ME (Analytical Model) and Simulation in QNAP-2 [22], it validates the credibility of ME solution.

In mobile ad-hoc networks the real traffic of a node is commonly concentrated in a small number of particular nodes. This characteristic has not been considered in the design of the existing routing algorithms. Therefore, it is difficult to guarantee performance in a simulation environment with realistic accuracy. To resolve this problem we propose a new routing algorithm called the Selective Route Discovery (SRD) algorithm. In this algorithm, each node selects frequently accessed nodes and periodically sends additional RREQ messages. Therefore, it can quickly adapt to the changes in network topology according to the movement of the nodes. This paper shows that the SRD algorithm has a shorter packet delivery time than the AODV algorithm when the simulation condition is improved so that the traffic concentration for each destination node varies.

Ad-hoc networks consist of only mobile nodes and have no support infrastructure. Due to the limited resources and frequent changes in topologies, ad-hoc network should consider these features for the provision of security. We present a lightweight secure routing protocol (LSRP) applicable for mobile ad-hoc networks. Since the LSRP uses an identity-based signcryption scheme, it can eliminate public or private key exchange, and can give savings in computation cost and communication overhead. LSRP is more computationally efficient than other RSA-based protocols because our protocol is based on the properties of pairings on elliptic curves. Empirical studies are conducted using NS-2 to evaluate the effectiveness of LSRP. The simulation results show that the LSRP is more efficient in terms of cost and overhead.

Almost every node of a Mobile Ad-hoc Network (MANET) has to perform the function of a router. The lifetime of participating nodes affects the stability of the network. Recent MANET routing protocols are greedy on network lifetime because of battery power limitations. Although, these algorithms help to maintain the stability of the network, they are not as much cost effective as traditional existing routing algorithms. Our proposed method considers both the routing cost and network lifetime issues in route selection, which is a good compromise between these two conflicting interests. The simulation results show that the proposed scheme selects a path with less cost than a path in lifetime prediction based routing algorithms and results more stable network than cost-effective routing algorithms do.

An ad hoc network has notable features such as frequent mobility, bandwidth limitation, and power constraints. Especially, due to the low bandwidth, there is a strong possibility to cause congestion. If there is congestion, however, power depletion and queuing delay will be serious problems in mobile nodes. In this paper, we propose a Weighted Load Aware Routing (WLAR) routing protocol, which shows excellent performance in an ad hoc network. WLAR Protocol, an extension of AODV, is to distribute the traffics among ad hoc nodes through load balancing mechanism. A new term of total traffic load is defined as a route selection metric, which is the product of average queue size and number of sharing nodes. Using NS2 simulator and real implementation, we show the performance, comparing to AODV.

The growing importance of Web traffic on the Internet makes it important that we have an accurate understanding of this traffic source in order to plan and provision. In this paper we present an empirical model of TCP connections used in delivering Web objects. Our TCP model takes advantage of a unique behavior of TCP that it alternates between inactive and active periods of transmitting data segments in bursts. Based upon the bursts in a TCP connection, we characterize TCP by defining the period between the starts of adjacent bursts and measuring the number of data segments transmitted and the time spent in this period. From the characterization we develop a TCP model that attempts to capture the major aspects of the real TCP connection.