تجزیه و تحلیل عملکرد از رفتار TCP در محیط های ماهواره ای جغرافیایی
کد مقاله | سال انتشار | تعداد صفحات مقاله انگلیسی |
---|---|---|
27554 | 2001 | 12 صفحه PDF |
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Computer Communications, Volume 24, Issue 9, 1 May 2001, Pages 877–888
چکیده انگلیسی
The paper shows the main problems that the Transmission Control Protocol (TCP) meets in a Geo Stationary Orbit (GEO) satellite environment characterized by high ‘delay per bandwidth product’. In a GEO environment, the delay in the delivery of a message is very high, the Round Trip Time (RTT) is above 500 ms. These characteristics heavily affect the acknowledgement mechanism on which the TCP is based and the performance of the protocol is much lower than in cable networks. The paper proposes and analyses possible solutions aimed at mitigating the negative effect and at improving the performance. A real test-bed is used to carry out the study: two remote hosts are connected together through a satellite link in the Ka-band (20–30 GHz) by using IP routers. The system has been tested by using a ftp-like application, that allows transferring data of variable size between the two hosts. This is due to the observation that most of the Internet multimedia applications, as browsing and distance learning, use massive file transfer. The protocol behavior is investigated both by tuning the parameters buffer size and initial congestion window and by modifying the dynamic characteristics of the slow start algorithm. The analysis itself allows suggesting solutions and taking decisions.
مقدمه انگلیسی
The opportunities offered by technologies like optical fibers, by the improved speed of devices and by the advanced network protocols allow the development of very complex multimedia applications. The new applications are designed not only for stand-alone PCs or workstations, but they need to run in an extended network environment. In this context a great interest has arisen in these last few years to connect Local Area Networks (LANs) by using satellite or terrestrial/satellite networks. The satellites have an inherent broadcast capability, they can connect remote sites when there is no terrestrial infrastructure, as in rural areas, and, at the same time, they can provide high-speed links. Due to their characteristics, they can represent an efficient way to provide efficient interconnections and multimedia services. Many national and international projects (listed extensively in Ref. [13]) in Europe, Japan and USA concern satellite networks and applications. In particular, some of them, or part of them, are aimed at improving performance at the transport level. NASA ACTS [8] and [11], ESA ARTES-3 [5] and Italian National Consortium for Telecommunications (CNIT)–Italian Space Agency (ASI) [1], which supports the present work, deserve specific attention, among many others. CNIT–ASI is a project aimed at analyzing the problems related to a satellite or terrestrial/satellite interconnection concerning both transmission and network problems. It is funded by the ASI, and carried out by the CNIT, a research center composed of several Italian universities. It is divided into two integrated lines: an experimental activity of multimedia services over a terrestrial-satellite network and a study activity for the system evolution concerning network protocols, integration of satellite and cable networks, medium access techniques, resource allocation, development of terminal equipment, and user interfaces. The project CNIT–ASI uses the ITALSAT satellite, works at 2 Mbits/s with an antenna of 1.8 m in the Ka-band (20/30 GHz). The paper focuses on a Geo Stationary Orbit (GEO) system with a large delay per bandwidth product and symmetric channel. Transmission Control Protocol (TCP) works quite efficiently over 64 or 128 kbytes GEO systems; the problem arises if the packets are transported at high speed and the network introduces a large latency. In a geo stationary system the Round Trip Time (RTT) is above 500 ms. The high delay to receive acknowledgements, on which the TCP is based, together with the large bandwidth used, makes the protocol inefficient [2] and the quality perceived by the users really poor. For instance, the delay of remote login or file retrieval in a satellite environment may be unacceptable for the user. On the other hand, in GEO systems there are also positive aspects: the RTT is approximately constant and the connectivity is guaranteed. The transmission errors measured are very low, at least in the test-bed used where error rates below 10−8 have been measured, and, as a consequence, the TCP does not fail to interpret each non-arrived packet as congestion presence. Ref. [4] contains a discussion about the TCP need to distinguish loss due to transmission errors from loss due to network congestion. The object of the paper is the investigation of the TCP behavior and the proposal of some modifications to improve performance. The topic has been investigated in the literature for some years: Ref. [15] contains a first overview. A more specific study in TCP/IP networks with high delay per bandwidth product and random loss may be found in Ref. [12]. More recently, Ref. [9] lists the issues and the challenges in satellite TCP and Ref. [10] highlights the ways in which latency and asymmetry impair TCP performance. Ref. [4] lists the main limitations of the TCP over satellite and proposes many possible methods to apply. A recent tutorial, which reports various improvements both at the transport level and at the application and network level, is Ref. [6]. This last paper focuses on the large delay per bandwidth product networks and suggests possible modifications to TCP, as the variation of the buffer size. The buffer size is also the object of the study [13], along with the initial congestion window. The solutions proposed in Ref. [13] are used also in this paper, which reports a new study about the effect of congestion and a new proposal of a modified version of the slow start algorithm. A preliminary version of the paper may be found in Ref. [14]. The modifications of the dynamic characteristics of the slow start algorithm are aimed at adapting the protocol to the characteristics of the channel. The performance analysis has been conducted experimentally by using a real test-bed composed of two hosts connected through a satellite link in the Ka-band (20–30 GHz). The paper is structured as follows. Section 2 contains a short description of the TCP congestion control characteristics. Section 3 introduces a possible parameterization of the protocol and two proposals to modify the slow start algorithm. The experimental environment is described in Section 4. Section 5 contains the results and the observations about the performance analysis. Section 6 reports the conclusions.
نتیجه گیری انگلیسی
The behavior of the TCP in a satellite link has been investigated. Parameters as the buffer dimension and the initial congestion window have been tuned. The dynamic of the slow start algorithm has been modified. The analysis itself has allowed suggesting some solutions. TCP results highly inefficient if used with no modification, the satellite imposes a delay and drastically reduces the performances of the protocol. The improvement due to an augmented dimension of the transmitter/receiver buffer and of the initial congestion window is outstanding, but a proper tuning of the two quantities to avoid congestion is strongly necessary. A large buffer, along with an extended initial window, makes the TCP more aggressive: on one hand, this improves the network throughput but, on the other hand, if the tuning is not precise, the risk of saturation is high. It is not simple to find a solution suited for any configuration. The impact of the link capacity and of the application environment is strong. An ‘ideal’ tuning may be found, as in the case treated, for a particular configuration. The modification of the slow start mechanism has been considered too. Even in this case, the window increase rate has to be ruled with great attention. If the non-modified scheme implies a low transfer, it is also true that a too quick increase creates congestion and, as a consequence, reduces performance. The system has been tested by using an ftp-like application, i.e. a file transfer application located just above the TCP, which has been taken as a reference, in a real satellite environment. An almost ideal channel has been measured and the lost packets have been substantially due to the saturation of intermediate routers. Nevertheless, a proper configuration scheme should be considered, particularly when there are multiple connections.