Modeling finite buffer effects on TCP traffic over an IEEE 802.11 infrastructure WLAN

doi:10.1016/j.comnet.2009.07.009

Computer Networks

Volume 53, Issue 16, 10 November 2009, Pages 2855-2869

https://doi.org/10.1016/j.comnet.2009.07.009 Get rights and content

Abstract

The network scenario is that of an infrastructure IEEE 802.11 WLAN with a single AP with which several stations (STAs) are associated. The AP has a finite size buffer for storing packets. In this scenario, we consider TCP-controlled upload and download file transfers between the STAs and a server on the wireline LAN (e.g., 100 Mbps Ethernet) to which the AP is connected. In such a situation, it is well known that because of packet losses due to finite buffers at the AP, upload file transfers obtain larger throughputs than download transfers. We provide an analytical model for estimating the upload and download throughputs as a function of the buffer size at the AP. We provide models for the undelayed and delayed ACK cases for a TCP that performs loss recovery only by timeout, and also for TCP Reno. The models are validated in comparison with NS2 simulations.

Introduction

We consider a scenario in which several clients or stations (STAs) are associated with a single Access Point (AP). The AP has a finite amount of FIFO buffer. For simplicity, we consider associations only at a single Physical (PHY) rate.¹ We are concerned with TCP-controlled file transfer throughputs when each STA is either downloading or uploading a single large file via the AP. The other endpoint of the transfers, or the “server,” is located on the high-speed Ethernet connected to the AP.² For such a situation, it has been reported that, with finite buffers at the AP, there is unfairness between the upload and download transfers with the upload transfers obtaining larger throughputs [1], [2]. Our objective is to provide analytical models that explain this unfairness, thus providing quantitative insights into the unfairness, and also predictive models for network engineering.

Relation to the Literature: Bruno et al. [3] analyzed the scenario of upload and download TCP-controlled file transfers in a single cell infrastructure WLAN when there is no packet loss at the AP. They assumed equal TCP windows for all connections, that the TCP receivers use undelayed ACKs, and showed that the total TCP throughput is independent of the number of STAs in the system; further the upload and download transfers each obtain an equal share of the aggregate throughput. A variation of this approach for modeling TCP transfers, along with a fixed-point analysis of 802.11e EDCA, was employed by Sri Harsha et al. [4] to provide a combined analytical model for TCP transfers, CBR packet voice, and streaming video over an infrastructure WLAN. The delayed ACK case was analyzed by Kuriakose et al. [5].

It is known that, if there is packet loss at the AP due to finite buffers, then in a situation of simultaneous upload and download transfers, the upload transfers each obtain a larger throughput than any of the download transfers. Gong et al. [1] provide simulation results validating this fact. They also show that as the AP’s buffer size increases, thus reducing packet loss at the AP, the throughput unfairness reduces. They also propose queue management strategies to alleviate the throughput unfairness. Pilosof et al. [2] analyzed the same problem of unfairness by assuming an M/M/1/K model for the finite buffer at the AP.

In this paper, we do not assume any conventional queueing model for the AP buffer, but we combine the earlier models for TCP-controlled file transfers in WLANs (i.e., [3], [5]) with a detailed model of TCP window evolution under tail-drop loss at the AP. A detailed modeling of TCP window evolution quantifying the unfair division of throughputs among the upload and download transfers providing valuable insights is the main contribution of this paper.

Outline of the Paper: The analytical model comprises two steps. In the first step (Sections 2 Throughputs: undelayed ACK, 3 Throughputs: delayed ACK), we use a simple extension of the analytical model of [3] to obtain the upload and download throughputs for a given value of h, the fraction of contention cycles in which the AP contends with a download packet (i.e., a TCP data packet) at the head-of-the-line (HOL) of its FIFO buffer. In the second step (Section 4), we obtain the value of h as a function of AP buffer size, using a detailed study of TCP window evolution when the upload connections have a maximum window limit but the download connections have no such window limit. We do this for both the undelayed and delayed ACK cases for the TCP version in which all loss recovery is by timeouts. We also provide a bound on h for the case when all the TCP connections have a maximum window limit. Simulation results that validate our analyses are provided in Section 6.

Section snippets

Throughputs: undelayed ACK

Consider an infrastructure mode WLAN with $N (= N_{d} + N_{u})$ STAs associated with the AP at the same PHY rate. Among these STAs, $N_{d}$ STAs each have a single download TCP connection while each of the remaining $N_{u}$ STAs have a single upload connection (see Fig. 1). All file transfers are to or from a “server” on the high-speed LAN to which the AP is connected. We analyze the case where TCP ACK transmissions on the WLAN use the “Basic Access” mode whereas TCP data transmissions use the “RTS–CTS” mode. This

Throughputs: delayed ACK

In the case of upload traffic with delayed ACKs, in steady state, every TCP ACK from the AP will generate two data packets at the STA. Thus, our earlier approximation that there can be at most one packet in the STA queue is no longer valid. However, assuming validity of the assumption that the transmission from the AP is always to an empty STA, we provide a simple upper bound on the throughput by assuming that whenever an STA wins the contention for the channel, it transmits both the TCP data

Determining h: TCP window analysis

In this section we obtain expressions for h for both delayed and undelayed ACK cases when the AP has a finite buffer, making suitable approximations in the process. Note that, as already stated earlier, our analysis is valid for the scenario when the STAs have TCP connections with a server located on the Ethernet to which the AP is connected, i.e., the delay between the server and the AP is negligible. In this section, the version of TCP analyzed does not support fast retransmit and fast

Extension to TCP Reno

The analysis easily extends to the Reno version of TCP. We first consider the undelayed ACK case, with no limit on the congestion window for download connections. The analysis is similar to that in Section 4.1. For the Reno case we assume that there are no timeouts and the recovery uses only Fast Retransmit and Fast Recovery mechanisms [11]. We assume that there are sufficient number of packets buffered for every download connection to trigger the Fast Retransmit mechanism. This will lead to

Analytical and simulation results

All the simulation results are obtained using ns-2.31 using parameters summarized in Table 3.

Conclusion

The analysis for calculating h is essentially rateless, i.e., the value of h does not change with PHY rate as long as the number of uploading STAs, the number downloading STAs, AP buffer size and maximum window limit for upload connections remain same. Thus, we can expect to obtain the same value of h even in the scenario where STAs associate with the AP with different PHY rates. The analysis for calculating h made use of only the fact that the average number of active STAs is small, as stated

Onkar Bhardwaj obtained his Masters in Telecommunication from the Indian Institute of Science, Bangalore, India in 2008 where he worked on IEEE 802.11 WLANs. His interests are Communication Networking and Algorithms. Currently he is with Computational Research Laboratories, Pune, India, where he is working on Numerical Linear Algebra algorithms.

References (11)

M. Gong, Q. Wu, C. Williamson, Queue management strategies to improve TCP fairness in IEEE 802.11 wireless LANs, in:...
S. Pilosof, R. Ramjee, D. Raz, Y. Shavit, P. Sinha, Understanding TCP Fairness Over Wireless LAN, in: Proceedings of...
R. Bruno, M. Conti, E. Gregori, Modeling TCP Throughput Over Wireless LANs, in: Proceedings of the 17th IMACS World...
S. Harsha, A. Kumar, V. Sharma, An analytical model for the capacity estimation of combined VoIP and TCP file transfers...
G. Kuriakose, S. Harsha, A. Kumar, V. Sharma, Analytical models for capacity estimation of IEEE 802.11 WLANs using DCF...

There are more references available in the full text version of this article.

Cited by (14)

A generic controller for managing TCP transfers in IEEE 802.11 infrastructure WLANs
2017, Journal of Network and Computer Applications
Citation Excerpt :
Another well-know QoS problem is that of unfairness between upload and download TCP-controlled bulk transfers reported in Wu et al. (2008). An analytical model that gives insight into this QoS problem was presented in Bhardwaj et al. (2009). Using this model, the authors in Bhardwaj et al. (2009) estimate the upload and download throughputs as a function of the buffer size at the AP.
In this paper, we present a generic controller that ensures fair and efficient operation of IEEE 802.11 infrastructure wireless local area networks (WLANs) with multiple co-channel access points. Our controller addresses performance issues of long-lived TCP transfers in multi-AP WLANs, by overlaying a coarse time-slicing scheduler on top of a cascaded fair queuing scheduler. The time slices and queue weights, used in our controller, are obtained from the solution of a constrained utility optimization formulation. A study of the impact of coarse time-slicing on TCP is also presented in this paper. We also present a methodology to improve the performance of co-existing short-lived and interactive TCP flows. Finally, we report the results of experiments performed on a real testbed, demonstrating the efficacy of our controller.
Aggregate download throughput for TCP-controlled long file transfers in a WLAN with multiple STA-AP association rates
2012, Performance Evaluation
Citation Excerpt :
However, our interest is in modeling aggregate TCP throughput, and the saturated traffic model does not capture the situation well. In the third category, Bharadwaj et al. [14] consider finite AP buffers, in contrast to the previous two, where AP buffers were assumed to be infinite; however, the single rate assumption is retained. The three groups mentioned above focus on long file transfers, where the TCP sender is assumed to have a file that is infinite in size.
We consider several WLAN stations associated at rates $r_{1}, r_{2}, \dots, r_{k}$ with an Access Point. Each station (STA) is downloading a long file from a local server, located on the LAN to which the Access Point (AP) is attached, using TCP. We assume that a TCP ACK will be produced after the reception of $d$ packets at an STA. We model these simultaneous TCP-controlled transfers using a semi-Markov process. Our analytical approach leads to a procedure to compute aggregate download, as well as per-STA throughputs, numerically, and the results match simulations very well.
Towards novel multipath data scheduling for future IoT systems: A survey
2021, arXiv
Fair Coexistence of Regular and Multipath TCP over Wireless Last-Miles
2019, IEEE Transactions on Mobile Computing
Modeling Compound TCP over WiFi for IoT
2018, IEEE/ACM Transactions on Networking
Analytical modeling of multipath TCP over last-mile wireless
2017, IEEE/ACM Transactions on Networking

View all citing articles on Scopus

G.V.V. Sharma was born in Visakhapatnam, India. He received the B.Tech. degree in Electronics and communication engineering from the Indian Institute of Technology, Guwahati, India, in 1999 and the M.Sc. (Eng.) degree in electrical communication engineering from the Indian Institute of Science, Bangalore, India, in 2004. Currently, he is pursuing the Ph.D. degree at the Department of Electrical Engineering, Indian Institute of Technology, Bombay, India. From August 2004 to July 2006, he was with the Applied Research Group of Satyam Computers, Bangalore. His research interests include communication theory and signal processing algorithms fro communication systems.

Manoj K. Panda obtained an M.Tech., degree in Electrical Engineering from Indian Institute of Technology (IIT) Kanpur, India, in March 2003. He was with the Applied Research Group (ARG), Satyam Computers Services Ltd., Bangalore, until January 2005 when he joined Indian Institute of Science (IISc) Bangalore as a Ph.D. student. He is currently working towards his Ph.D. His areas of interest include modeling, model based simulation and optimization of communication networks, in general, and of wireless local area networks, in particular.

Anurag Kumar (B.Tech., IIT Kanpur, Ph.D. Cornell University, both in EE) was with Bell Labs, Holmdel, for over 6 years. He is now a Professor in the ECE Department at the Indian Institute of Science (IISc), Bangalore, and Chair of the Division of Electrical Sciences. His area of research is communication networking, and he has recently focused primarily on wireless networking. He is a Fellow of the IEEE, of the Indian National Science Academy (INSA), and of the Indian National Academy of Engineering (INAE). He has been an associate editor of IEEE Transactions on Networking, and of IEEE Communications Surveys and Tutorials. He is a coauthor of the advanced text-books “Communication Networking: An Analytical Approach,” and “Wireless Networking,” by Kumar, Majunath and Kuri, published by Morgan-Kaufman/Elsevier.

^☆: This is an expanded version of a paper that appeared in COMSNETS 2009 (The First International Conference on COMmunication Systems and NETworkS), Bangalore, India, January 5–10, 2009. The research on which this paper is based was supported by the Department of Information Technology, Government of India, and Airtight Networks, Pune, India.

View full text

Modeling finite buffer effects on TCP traffic over an IEEE 802.11 infrastructure WLAN☆