This paper studies the transmission control protocol (TCP) performance over a discrete multi-tone (DMT) based asymmetric digital subscriber loop (ADSL) network. The impact of DMT subchannel bit loading on the TCP throughput performance is studied. The simulation results show that there is a threshold for the signal-to-noise ratio (SNR) gap or bit error rate (BER) above which TCP throughput drops quickly. This threshold takes its value in a wide range depending on the TCP round-trip time as well as channel noises. This suggests that it would be insufficient to set a fixed target BER at, e.g. 10−7, when calculating the number of bits to be loaded in each subchannels. Instead, the bit loading should take TCP performance into account. Finally a dynamic bit loading scheme is proposed, which jointly optimizes the channel bit rate and TCP throughput performance.
As an emerging technology, asymmetric digital subscriber loop (ADSL) has attracted a lot of attention due to its ability to deliver broadband access over the traditional telephone lines. However, unlike a backbone network based on fiber optics technology with a bit error rate (BER) on the order of 10−11–10−13, an ADSL system has to live with potentially high and variable BER, ranging from 10−3 to 10−9. Except the ADSL channel characteristics, four major noises contribute to the high and variable BER, including white Gaussian noise, far end crosstalk (FEXT), near end crosstalk (NEXT), and impulse noise.
Tremendous research efforts have been made in the analysis of the impact of channel noises on the ADSL performance, e.g.[10] and [15], and in the design of loading and dynamic loading algorithms to optimize the channel performance [7] and [9]. Of particular interest is the rate-adaptive (RA) loading algorithm [9] for ADSL systems based on discrete multi-tone (DMT) modulation. The RA loading algorithm maximizes the overall bit rate subject to a fixed energy constraint and signal-to-noise ratio (SNR) gap or BER.
Most of the data applications are built on top of the transmission control protocol (TCP). Therefore, in parallel to the above development, research effort has been made on the study of the performance of TCP over asymmetric and lossy channels [1], [14] and [16]. By assuming that the channel bit rates in both directions are given, these papers investigate the effects of buffering [16], asymmetry [4], and random loss or BER [5] and [18] on TCP throughput performance.
However, none of the studies mentioned above considered both the physical layer (ADSL) and the upper layer (TCP) performance simultaneously. Existing TCP performance papers mentioned above assume the underlying channel conditions are given. For instance, by setting BER at 10−7 and the maximum bit rates at 8 Mbps downstream and 800 kbps upstream, respectively, the TCP performance can then be independently evaluated regardless of the actual underlying ADSL processes. In reality, however, the maximum bit rates are complicated functions of BER as well as subchannel SNRs, which may change from time to time. Hence, analyzing TCP performance over ADSL should take the physical channel processes into account. On the other hand, the research on the loading and dynamic loading algorithm design for DMT modulation did not take higher layer performance into account. Although it is mentioned in Ref. [9] that some higher-layer entity may arbitrate when the reloading should occur for dynamic loading, the question as to which higher layer entity and how a higher layer entity makes the reloading decision is not addressed. Since the objective of performance optimization at any given layer is to deliver the best service to its upper layer, TCP performance should be the ultimate performance measure for applications using TCP as their underlying protocol.
In this paper, the performance analysis of TCP over ADSL is performed by taking into account of the underlying ADSL processes. This approach enables us to find the true TCP performance limit for different standard ADSL test loops and under various noise conditions. The performance analysis further leads to the development of a bit loading scheme which jointly optimizes channel bit rate and TCP throughput performance. The idea is first to take BER as a variable, rather than a fixed target, and run the RA bit loading algorithm [9] to find the functional relationship between the maximum bit rate and BER. Then locate an operating point on the curve of the maximum bit rate versus BER, which maximizes the TCP throughput performance. Finally, load the subchannels with the energies or numbers of bits per symbol calculated at this operating point. Obviously, at this operating point, both physical layer and TCP throughput performance are jointly optimized.
The rest of the paper is organized as follows. Section 2 presents a background introduction to DMT based ADSL systems. Section 3 presents the performance evaluation of TCP performance on two test loops of a ADSL system. Section 4 describes a joint optimization scheme to maximize the TCP performance. Finally, Section 5 concludes the paper and presents a future research direction.
In this paper, the performance of TCP over DMT based ADSL is studied. In contrast to the previous papers, this study did not assume that the underlying ADSL channel conditions are fixed. Instead, it explicitly considered the impact of ADSL channel conditions on the TCP throughput performance. This study enabled us to identify the limit of TCP throughput performance under different standard ADSL test loops and various noise conditions. This study further lead to the design of a dynamic bit loading scheme which jointly optimizes the channel bit rate and TCP throughput performance.
However, the study only considered the TCP throughput as a performance measure. A possible extension of this work is to consider multiple classes of service. Results on providing multiple classes of service on the ADSL transmission line are available [11] and [12]. Our future work is to address the issue as to how the physical layer service differentiation is to be related to the higher layer service differentiation.
