Authors:

Jingyuan Wang (wangjingyuan06@mails.tsinghua.edu.cn), Jiangtao Wen (jtwen@tsinghua.edu.cn), Jun Zhang, Yuxing Han (ericahan@ee.ucla.edu)

Papers:

Jingyuan Wang, Jiangtao Wen, Jun Zhang and Yuxing Han, "TCP-FIT - an improved TCP congestion control algorithm and its performance", Proc. 2011 IEEE Infocom, Shanghai, October, 2011.

Jingyuan Wang, Jiangtao Wen, Jun Zhang, Yuxing Han, "TCP-FIT - a novel TCP congestion control algorithm for wireless networks", IEEE Globecome 2010 Workshop on Advances in Communications and Networks, December, 2010, Miami, FL.

Jingyuan Wang, Jiangtao Wen, Jun Zhang, Yuxing Han, "A demonstration of a new TCP congestion control algorithm over LTE and other challenging networks," MobiCom 2010 Demo, Sept. Chicago, IL. (PDF)


Introduction

TCP-FIT is a novel congestion control algorithm targeting both emerging wireless networks such as LTE, WiMax, Wi-Fi and HSPA and high speed long delay (high BDP) networks.

The Transmission Control Protocol (TCP) is a reliable transport layer protocol that is widely used on the Internet. Congestion control algorithm is an integral module of TCP that directly determines the performance of the protocol. Standard congestion control algorithms such as TCP-Reno and TCP-NewReno, achieved great success for several decades but are found to perform poorly over wireless and/or high Bandwidth Delay Product (BDP) links. To improve TCP performance over wireless and high BDP networks, many TCP variants have been proposed, including TCP Westwood, TCP Veno for wireless applications and Compound TCP, TCP CUBIC, FAST TCP for high BDP networks. Although these algorithms have achieved success in their respective target applications, designing a TCP congestion control algorithm that perform gracefully in both wireless and high BDP networks is still a great challenge. On the other hand however, with the deployment of wireless networks, such as LTE, WiMAX, as well as high bandwidth, real time applications such as multimedia over TCP/HTTP, it is required for the TCP congestion algorithm to handle both wireless connections with random radio related losses as well as congestion-introduced issues typical for wired high BDP networks.

To address these problems, a novel TCP congestion control algorithm for the heterogeneous networks which contained high BDP links and wireless links, named TCP-FIT, is proposed. A demo comparison between TCP-FIT and existing TCP algorithms over an wireless network is shown in the following video.

As can be seen in fig. 1 to fig. 8, given the same challenging network conditions, TCP-FIT achieves significant performance improvement as compared with other algorithms.

 

 

Fig.1 Average throughput of different TCP algorithms over CHINA TELECOM CDMA 2000 3G network

 

Fig.2 Throughput variation of of different TCP algorithms over CHINA TELECOM CDMA 2000 3G network in 2 hours

 

 

Fig.3 Average throughput of different TCP algorithms over WIFI network in Tsinghua University

 

Fig.4 Throughput variation of different TCP algorithms over WIFI network in Tsinghua University in 2 hours

 

Fig.5 Average throughput of different TCP algorithms over WIFI network at a Starbucks coffee shop in Beijing

 

Fig.6Throughput variation of different TCP algorithms over WIFI network at a Starbucks coffee shop in Beijing in 2 hours

Fig.7 Average throughput of different TCP algorithms over LTE network

 

Fig.8 Throughput comparison between different TCP algorithms for varied LTE UE SINR.

To test the performance of TCP-FIT over live, deployed, real-world networks, we put a TCP-FIT server, which located Orange, California, US, on the Internet and tested TCP-FIT using clients located in 5 different cities/towns in 4 countries (China, Switzerland, USA and India) on 3 continents. At each location, we tested a combination of wired line connections, Wi-Fi and whenever possible, 3G wireless networks. The location, network and OS of the test clients are listed in Table IV. In each experiment, a script was used to automatically and periodically cycle through different TCP algorithms on the server over long durations of time (4-24 hours), while the client collected throughput information and other useful data. The period for changing the TCP algorithms was set to about 5-10 minutes, so that 1) the algorithms tested were able to reach close to steady-state performances; 2) the period is consistent with the durations of a reasonably large percentage of the TCP based sessions on the Internet (e.g. YouTube streaming of a single piece of content, synchronizing emails, refreshing one web page, etc.). In the performance measure, TCP-FIT are compared with CUBIC, CTCP, Reno in all cases. HSTCP and Illinois are compared for wired network and TCPW, Veno for wireless. The location, network and OS of the test clients are listed in following table:

Figures

Location

Network

Client OS

Figure 9

Zurich

Ethernet

Win Vista

Figure 10

LA

Ethernet

Win Vista

Figure 11

Beijing

ADSL

Win XP

Figure 12

Zurich

WIFI

MAC OS

Figure 13

LA

WIFI

Win Vista

Figure 14

Beijing

WIFI

Linux

Figure 15

Beijing

CDMA 2000

Win XP

Figure 16

Fujian

CDMA 2000

Win Vista

Figure 17

Bangalore

ADSL

Win Vista

Our results are summarized in Following Figures:
 

Figure 9. Wired Network in Zurich

Figure 10. Wired Network in Los Angeles
 

Figure 11. Wired Network in Beijing

 

Figure 12. Wireless Network in Zurich

 

Figure 13. Wireless Network in Los Angeles

 

Figure 14. Wireless Network in Beijing

 

Figure 15. 3G Network in Beijing

 

Figure 16. 3G Network in a town of Fujian Province

   

Figure 17. ADSL Network in Bangalore

   
we compared the performance of TCP-FIT with widely deployed TCP variants, TCP Reno and CUBIC, using the Planet-Lab testbed. In the experiments, 245 nodes of the Planet-Lab test bed distributed over 192 cities in 43 countries were used to download video clips from an HTTP servers located in San Diego, CA. The geographical distribution of these nodes is shown in Figure 18. These nodes covered 233 ISPs representative of the conditions of the current Internet.
Figure 18. Geographical Distribution of Nodes in Planet-lab Experiments
The cumulative probabilistic function for the download throughputs of the nodes is shown in Figure 19, while the corresponding cumulative probabilistic function for the TCP-FIT to Reno/CUBIC speedup is in Figure 20.
 
Figure. 19 Cumulative Probabilistic Function for Throughputs
 
Figure. 20 Cumulative Probabilistic Function for Speedup

Lab of Multimedia and Networking
Institute of HCI and Media Integration
Department of Computer Science and Technology
Tsinghua University

Mail: tcpfit.thu@gmail.com

Address: Room 1-512, FIT Building , Tsinghua University, Beijing 100084, P.R.China

 

tumblr visitor