Performance Evaluation of TCP Vegas Versus Different TCP Variants in Homogeneous and Heterogeneous Networks by Using Network Simulator 2

International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 90

Performance Evaluation of TCP Vegas versus Different TCP Variants in Homogeneous and Heterogeneous Networks by Using Network Simulator 2

First B. S. Yew, Second B. L. Ong, and Third R. B. Ahmad

 congestion due to packet losses and unusual delay. There is no Abstract— The performances of TCP Vegas versus different algorithm that is designed to minimize the transmission loss as TCP variants in homogeneous and heterogeneous networks were the transmission loss in a wired network is negligible. In evaluated via simulation experiment using network simulator wireless network, the performance of TCP is poor. This is due (ns-2). In our simulation experiments, we investigate the to the transmission losses in the network [3]. The transmission performance of TCP Vegas in both homogeneous and losses in wireless network are not negligible. Transmission heterogeneous wired and wired-cum-wireless networks. In losses in wireless network refer to the high bit error, noise homogeneous wired network, TCP Vegas outperforms other TCP variants. In contrast, the performance of TCP Vegas is the interference and channel errors during the transmission of data worst compared to other TCP variants in homogeneous from the source network to the receiver network. TCP treats wired-cum-wireless network. For both heterogeneous wired the transmission loss as a signal of congestion and invokes its network and heterogeneous wired-cum-wireless network, TCP congestion control. When TCP congestion control responds to Vegas is unable to achieve fair bandwidth allocation when these transmission losses, the assumptions in [2] are violates sharing the bottleneck link with other TCP variants. On the and degrade the end-to-end performance of TCP. other hand, TCP Vegas has a significantly lower delay than other In our research, the performances of TCP Vegas in TCP variants in both homogeneous and heterogeneous wired and wired and wireless network were analyzed. However, instead wireless networks. of implementing TCP Vegas in a purely wireless network that involves the sender and receiver networks as the wireless Index Term— TCP Vegas, Homogeneous, Heterogeneous, environment, we implement wired-cum-wireless network in Wired Network, Wired-cum-Wireless Network. our simulation experiment. The wireless network is chosen as the source (sender) network while the destination (receiver) network is the wired network. We aim to analyze the performance of TCP Vegas as the TCP sender agent in the I. INTRODUCTION wireless network. TCP Vegas is the alternative TCP congestion control algorithm is designed as a mechanism to minimize congestion losses is a wired network [1]. The implementation of TCP congestion control that only involves performances of TCP as a reliable end-to-end data transfer is modification on sending side, proposed by [4].Thus, only the optimizes in wired network due to the implementation of TCP sender network is set to the wireless network while the congestion control algorithm. However, when TCP congestion receiver network is set to wired network with the TCP receiver control algorithm is implemented in wireless network, the agent set to be TCP sink. performance of TCP degrades. The performance of TCP Vegas is compared against In Wired network, congestions in the network occur due other TCP variants. For all TCP variants included TCP Vegas, to packet losses and packet delay. TCP congestion control the performance in wired-cum-wireless network is worst compared to their respective performance in wired network. assumptions in [2] only involve the algorithms to control For TCP Vegas that is implemented in homogeneous network, its performance outperforms other TCP variants in wired network but perform the worst in the wireless network. This is First-B. S. Yew is a student conducting the research on enhancing the performance of TCP Vegas in computer network, University Malaysia Perlis due to the delay estimation scheme of TCP Vegas congestion (e-mail: [email protected]). control algorithm. In wireless network, it is difficult to obtain Second- B. L. Ong is a senior lecturer (PhD) of the Computer and precise delay estimation. Wireless network consist of mobile Communication Engineering Department, University Malaysia Perlis, nodes moving in random direction. These random movements Malaysia. Her research interest includes Wireless Communication, Simulation using ns-2 @ omnett, Mobility Management, Session Initiation Protocol reduce the accuracy of estimation in round trip time and affect (SIP), Cellular Network, IPV6 Network, and Wireless Mesh Network (WMN) the efficiency of TCP Vegas congestion control algorithm in (e-mail: drlynn@ unimap.edu.my). wireless network. The absence of rerouting detection

Third-R. B. Ahmad is the dean (PhD, Associate Professor) of the Computer mechanism in TCP Vegas congestion control algorithm also and Communication Engineering Department, University Malaysia Perlis, Malaysia. His research interest includes Computer, Telecommunication and affects its performance in mobile environment. TCP Vegas is Optical Network Modeling, Embedded System Applications Using Single unable to differentiate the increment of delay is due to Board Computer and GNU/Linux (e-mail: [email protected]).

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 91 rerouting or packet losses and treats all of delay increment as quantifying the joint effect of parameters of automatic repeat packet losses and invokes its congestion control algorithm. As request (ARQ)/ forward error connection (FEC). a result, the sending rate of TCP Vegas is reduces and its K. Pentikousis et. al [10] conduct a survey study on the performance degrades [5]. For TCP Vegas that is implemented performance of the TCP in wired-cum-wireless environments in heterogeneous networks, it is unable to achieve fair From their survey, the present TCP performance-related issues bandwidth allocation in both wired and wireless network. This in wired-cum-wireless environments. They conclude that TCP is mainly due to TCP Vegas proactive congestion control perform poorly in wireless environments in terms of achieved algorithm that tries to prevent incipient congestion by reducing throughput due to the factors of limited bandwidth, Long RTT, its sending rate and resulted in unfair domination of available random losses, short flows, user mobility and power bandwidth [6][7]. consumption. This paper is organized as follows. In section 2, we did H. Balakrishnan et. al [11] compare the mechanisms for literature review on previous works on TCP Vegas improving TCP performance over wireless links. They performances. In section 3, we explain the simulation setup classified these mechanisms into three categories. The first that is used to conduct this simulation experiment using category is the end-to-end protocol, where loss recovery is network simulator (ns-2). The simulation results that are performed by the sender. The second category is the link-layer obtained are discussed in section 4. In section 5, we discuss on protocols that provide local reliability; and the third category the future work of the simulation experiment. Finally, in is the split-connection protocols that break the end-to-end section 6, we conclude on the performances of TCP Vegas on connection into two parts at the base station. These protocols both homogeneous and heterogeneous in wired-cum-wireless shield the wireless transmission losses from the congestion network. losses.

II. PREVIOUS WORK III. SIMULATION SETUP L. S. Brakmo et. al [3] had proposed a new technique of The simulation tool used for this simulation experiment is congestion control detection and avoidance called TCP Vegas. ns-2. The simulation environments were divided into two They investigate the performance of TCP Vegas in the wired parts: Homogeneous and heterogeneous wired networks, and network by comparing to the performance of TCP Reno. The Homogeneous and heterogeneous wired-cum-wireless implementation of BSD Unix in the x-kernel based simulator networks. tool is used to analyze the performance of TCP Vegas. From their study, they conclude that TCP Vegas achieves better A. Homogeneous and heterogeneous wired networks throughput. TCP Vegas involves modification to the earlier TCP congestion control algorithm (TCP Tahoe and TCP Reno The network topology used in the simulation [8]) that detects congestion based on packet delay rather than experiment is shown in Fig. 1. The network topology used is a packet loss. The delay estimation scheme is used as a simple dumbbell topology which consists of TCP senders, mechanism to detect the congestion in terms of packet delay. TCP receivers and a pair of routers. The links between the TCP TCP Vegas as the congestion control algorithm optimizes the senders and router 1 are called as the sender links while the performance of TCP due to the implementation of this delay links between the TCP receivers and the router 2 are called the estimation scheme. However, this is only true when TCP receiver links. The sender and receiver link represent a local Vegas is implements in homogeneous wired network. When area network (LAN). The link between router 1 and router 2 is TCP Vegas is implemented in heterogeneous wired network, called the bottleneck link. The bottleneck link represents a TCP Vegas performances become worst. wide area network (WAN). For Homogeneous and A. D. Vendictis et. al [6] evaluate the performance of heterogeneous wired networks, we set four simulation TCP Vegas and TCP Reno behaviors in a heterogeneous wired environments as shown in Table I. network. From their study, they conclude that the fairness of TCP Vegas and TCP Reno cannot be achieved. They had proposed a new TCP congestion control algorithm called TCP NewVegas. TCP NewVegas preserves the excellent features of TCP Vegas in homogeneous wired network and improves the bandwidth fairness of bottleneck link when competing with TCP Reno. R. Dunaytsev et. al [9] evaluate the performance of TCP in wired and wired-cum-wireless networks. They proposed enhanced-TCP variants that allow evaluating the combined effect of parameters on TCP performance over both correlated and uncorrelated wireless network. These enhanced-TCP models include TCP Tahoe, TCP Reno, TCP NewReno and TCP SACK. The parameters include the bit error rate (BER). The developed models are able to optimize startup Fig. 1. Dumbbell wired network topology performance, minimize correlated losses, enable performance evaluation over wide range of operating conditions and allow

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 92

The sender links and the receiver links are a full wired duplex link. The bandwidth of the sender links that implement TABLEIV Ethernet LAN is 10Mbps and the link delay in the full duplex HETEROGENEOUS WIRED NETWORK TCP TCP Case Ethernet LAN is set to 10ms. The bandwidth of the sender Sender A Sender B links that implement Fast Ethernet LAN is 100Mbps and the 1 Vegas Tahoe link delay of in the full duplex wired Fast Ethernet LAN is set to 1ms. The bottleneck link is a full wired duplex link with the 2 Vegas Reno capacity of 2Mbps that represents current bandwidth of 3 Vegas NewReno MyREN network. The link delay of the bottleneck link is set to 4 Vegas SACK 50ms. The simulation parameters of the network topology are showed in Table II. In homogeneous network, both TCP sender implements the B. Homogeneous and heterogeneous wired-cum-wireless same TCP source. The TCP source used on the sender side in networks the homogeneous network is varied from Tahoe, Reno, The network topology used in the homogeneous and NewReno, Vegas and SACK. For the receiver side, we set heterogeneous wired-cum-wireless networks simulation TCP Sink as the TCP source. There are five cases considered experiment is shown in Fig.2. The network topology used is in simulation experiment of homogeneous wired network as the similar as in wired network, which is the simple dumbbell shown in Table III. The performance of TCP variants in topology. However, in wireless-cum-wired network, base homogeneous network is evaluated in order to analyze the station is added. The base station is the wireless access point average throughput and average delay. which connect TCP sender agent in the sender network to the In heterogeneous network, we set TCP Vegas as the TCP wireless carrier network. source on the sender side of the flow A. For flow B, the TCP source on the sender side is varied. There are four cases considered in the simulation experiments of heterogeneous wired network as shown in Table IV. The performance of TCP Vegas versus other TCP variants is evaluated in order to investigate the throughput fairness and average delay.

TABLEI SIMULATION ENVIRONMENTS IN HOMOGENEOUS AND HETEROGENEOUS WIRED NETWORKS Simulation TCP Sender TCP Receiver Environment (A&B) (A&B) Fig. 2 Dumbbell wired-cum-wireless topology (TCP sender = wireless 1 Ethernet Ethernet network, TCP receiver = wired network)

2 Ethernet Fast Ethernet In wired-cum-wireless network, the IEEE 802.11b wireless local area network (WLAN) is implemented in the 3 Fast Ethernet Fast Ethernet sender network. We implemented the physical specifications 4 Fast Ethernet Ethernet of a wireless ORiNOCO 11b Client PC card in an open range which covered the transmission range of 160m. The data rate of IEEE 802.11b WLAN is 11 Mbps with the basic rate 1Mbps TABLEII SIMULATION PARAMETERS and operated at frequency of 2.4GHz. From the data sheet of the ORiNOCO 11b Client PC card, we obtain the respective Link Bandwidth Delay value of the parameters that is implemented in our simulation experiment. Bottleneck 2 Mbps 50ms For the receiver network, wired fast Ethernet LAN Ethernet 10 Mbps 10ms network is implemented. The receiver link is a full wired duplex link. The bandwidth of the fast Ethernet LAN is Fast Ethernet 100 Mbps 1ms 100Mbps and the link delay in the full duplex Ethernet LAN is set to 1ms. The bottleneck link is a full wired duplex link with TABLEIII the capacity of 2Mbps that represents current bandwidth of HOMOGENEOUS WIRED NETWORK TCP TCP MyREN network. The link delay of the bottleneck link is set to Case Sender A Sender B 50ms. The simulation parameters of the wired-cum-wireless 1 Vegas Vegas networks are showed in Table V. For both homogeneous and heterogeneous 2 Tahoe Tahoe wired-cum-wireless networks, we consider simulation 3 Reno Reno environments which include the wireless network as the 4 NewReno NewReno sender network while the wired network as the receiver 5 SACK SACK network. In homogeneous wired-cum-wireless network, TCP senders implement the same TCP source. In heterogeneous

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 93 wired wired-cum-wireless network, TCP senders implement network [5]. This is because the estimation of RTT by using different TCP source and share the available bandwidth. TCP fine-grained times considers the timestamp and the timeout of receiver agent is set to TCP sink in both homogeneous and packet that is sent from the sender to the receiver. TCP Vegas heterogeneous wired-cum-wireless networks. The simulation records the timestamp of the respective packet in a network. environment for both homogeneous and heterogeneous There are two situations considered, that is when duplicate and networks is shown in Table VI. non duplicate acknowledgements (ACK) are received at the sender side. When duplicate ACKs are received at the sender TABLEV side, TCP Vegas checks the difference of the current time SIMULATION PARAMETERS OF WIRED-CUM-WIRELESS NETWORK interval and the timestamp recorded. As the difference is larger Link Bandwidth Delay than the timeout, TCP Vegas retransmits the lost packet Wireless Network 11Mbps without having to wait for the third duplicate ACK that are 1us (Sender) transmitted by the receiver to the sender. TCP Vegas also Wired Network detect any other packets that have been lost previously by 100 Mbps 1ms (Receiver) retransmitting the respective packet [1]. This is when the sender received a non-duplicate ACK. As the time interval of Bottleneck 2 Mbps 50ms the last packet that was sent to the receiver side is larger than the timeout value, TCP Vegas retransmits the respective TABLEVI packet. The retransmission mechanism of TCP Vegas SIMULATION ENVIRONMENT OF HOMOGENEOUS AND HETEROGENEOUS minimizes the packet losses in a network. As the packet losses WIRED-CUM-WIRELESS NETWORK in a network are minimized, the resulted average throughput of Homogeneous Heterogeneous TCP Vegas is optimized. Hence, this concludes why the average throughput of TCP Vegas outperforms other TCP wired-cum-wireless network wired-cum-wireless network Case variants. TCP Sender TCP Sender As for the average delay, the RTT estimation in TCP Vegas congestion control algorithm enables the detection of A B A B congestion in a network at early stage. This means that TCP 1 Vegas Vegas Vegas Tahoe Vegas detects congestion faster than other TCP variants. Early stage detection of congestion reduces the delay in a network. 2 Tahoe Tahoe Vegas Reno This is because the network does not have to waste time on 3 Reno Reno Vegas NewReno waiting for a packet lost to conclude that congestion occurs in a network. This explains why the average delay in a network 4 NewReno NewReno Vegas SACK that is implementing TCP Vegas source is less. The simulation 5 SACK SACK result in the homogeneous wired network is tabulated in Table VII and refers to the appendix for the graphical results (Fig.A1 and Fig. A2). IV. RESULT AND DISCUSSION TABLEVII SIMULATION RESULT IN HOMOGENEOUS WIRED NETWORK In this section, we present the simulation results and the Fast analysis for both homogeneous and heterogeneous wired and Ethernet/ Fast wired-cum-wireless network. Simulation Ethernet/ Ethernet/ Fast Ethernet/ Environme Ethernet Fast A. Wired network Ethernet Ethernet nt (E/E) Ethernet (i) Homogeneous Wired Network (E/F) (F/E) For all the simulation environments in the homogeneous (F/F) wired network, TCP Vegas performs better than other TCP Average Throughput (kbps) variants. The average throughput of TCP Vegas outperforms Vegas 1983.240 1986.000 1988.400 1986.000 the other four TCP variants. The average delay experience by TCP Vegas is the lowest among the other four TCP variants. Tahoe 1900.613 1910.099 1925.499 1910.099 The lower the delay, the faster the packet of data can be sent to Reno 1964.554 1968.373 1972.069 1968.373 the destination, this improves the efficiency of the network. NewReno 1971.822 1975.642 1980.816 1975.378 Here, we discuss the reasons which contribute to the better performance of TCP Vegas. SACK 1822.504 1830.388 1831.867 1831.744 TCP Vegas congestion control algorithm works based on Average Delay (Second) the estimation of round trip time (RTT). TCP Vegas implements fine-grained timer in RTT estimation of the Vegas 0.205 0.190 0.136 0.190 retransmission mechanism. The accuracy of the fine-grained Tahoe 0.554 0.543 0.518 0.544 timer is higher than the coarse-grained timer that is Reno 0.571 0.560 0.542 0.559 implemented in TCP Tahoe, TCP Reno, TCP NewReno and SACK. The fine-grained timer in RTT estimation of the NewReno 0.570 0.560 0.540 0.560 retransmission mechanism minimizes the packet losses in a SACK 0.552 0.536 0.512 0.539

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 94

(ii) Heterogeneous Wired Network TABLEVIII RESULT OF TCP VARIANTS IN HETEROGENEOUS WIRED NETWORK In heterogeneous wired network, the performance of TCP Fast Ethernet/ Fast Vegas against TCP variants in a network where the bottleneck Ethernet/ Ethernet/ Simulation Fast Ethernet/ Ethernet Fast link is shared Is evaluated. The average throughput fairness of Environment Ethernet Ethernet (E/E) Ethernet TCP Vegas when sharing the bottleneck link with other TCP (E/F) (F/E) (F/F) variants is analyzed. Average Throughput (kbps) For all the simulation experiments in the heterogeneous Vegas / 218.640 / 80.640 / 218.16 / 80.640 / wired network, TCP Vegas is unable to achieve fair bandwidth Tahoe 1705.707 1835.190 1722.34 1835.190 allocation when sharing the bottleneck link with different TCP Vegas / 191.640 / 77.280 / 152.88 / 77.280 / Variants. TCP Vegas receives unfair average throughput Reno 1778.765 1888.906 1819.42 1888.906 compared with TCP Tahoe, TCP Reno, TCP NewReno and Vegas / 190.320 / 77.280 / 88.20 / 77.280 / TCP SACK. TCP Vegas received the most unfair throughput NewReno 1783.077 1893.834 1896.17 1893.834 when it is sharing the bottleneck with TCP Reno and TCP Vegas / 246.360/ 84.000 / 134.280 / 84.000 / NewReno. (TCP NewReno is a slight modification over TCP SACK 11646.694 1790.099 1756.589 1790.099 Reno. The congestion mechanisms implemented in TCP Average Throughput Fairness (%) NewReno are slightly the same as TCP Reno except that TCP Vegas / 10.932 / 4.032 / 10.908 / 4.032 / NewReno is able to detect multiple packet losses. Hence the Tahoe 85.256 91.760 86.115 91.760 results of TCP Reno and TCP NewReno do not vary much.) Vegas / 9.582 / 3.864 / 7.644 / 3.864 / TCP Reno and TCP NewReno dominate most of the available Reno 88.938 94.445 90.971 94.445 bandwidth of the bottleneck link and being the most unfair to Vegas / 9.516 / 3.684 / 4.410 / 3.684 / TCP Vegas. This is because the congestion avoidance NewReno 89.154 94.692 94.809 94.692 mechanism of TCP Reno and NewReno is aggressive Vegas / 12.318 / 4.200 / 6.714 / 4.200 / compared with TCP Vegas. In order to fully utilize the SACK 82.335 89.510 87.830 89.510 available bandwidth of the bottleneck link, TCP Reno and Average Delay (Second) TCP NewReno continue to increase the window size until Vegas / 0.249 / 0.242 / 0.220 / 0.242 / Tahoe 0.284 0.273 0.265 0.273 packet losses occur. As window size increases, the buffer size Vegas / of TCP Reno and TCP NewReno in the bottleneck link also 0.258 / 0.256 / 0.241 / 0.256 / Reno 0.285 0.276 0.266 0.276 increases. The larger buffer size keeps more packets and hence Vegas / 0.260 / 0.256 / 0.255 / 0.256 / dominates most of the available bandwidth in the bottleneck NewReno 0.285 0.276 0.269 0.276 link. Thus, TCP Reno and TCP NewReno connections Vegas / 0.231 / 0.231 / 0.218 / 0.231 received higher throughput as the buffer size is larger. On the SACK 0.284 0.275 0.265 /0.275 other hand, TCP Vegas congestion avoidance mechanisms detect congestion at early stage. TCP Vegas detects congestion faster than TCP Reno and TCP NewReno. TCP Vegas B. Wired-Cum-Wireless Network congestion avoidance mechanism reduces the window size in (i) Homogeneous Wired-Cum-Wireless Network order to maintain a smaller buffer queue size. The smaller buffer size minimizes the packet losses due to buffer For all the simulation environments in the homogeneous overflows [8]. Thus, the TCP Vegas uses smaller bandwidth wired-cum-wireless network, TCP Vegas performance is the capacity in the bottleneck link. This concludes why TCP worst compared to other TCP variants. TCP NewReno Vegas is unable to receive fair bandwidth allocation and the achieves better throughput than other TCP variants. On the resulted throughput is significant low when sharing the other hand, the average delay experience by TCP Vegas is bottleneck link with TCP Reno, TCP NewReno and other TCP significantly lower than the other four TCP variants. variants. Also, from all the simulation results, the value of TCP Vegas controls network congestion by using delay delay experiences in the connection that is implementing TCP estimation (RTT). In wireless network that consist mobile Vegas is smaller. This is due to the smaller buffer size that is nodes, packet losses occur due to node movements and maintained by TCP Vegas in the bottleneck link. The smaller wireless channel errors. Due to the random movement of buffer size resulted in smaller value of delay. On the other mobile nodes, it is difficult for TCP Vegas to obtain accurate hand, in the connection that is implementing other TCP estimation on the RTT estimation to adjust its window size in a variants, the buffer size is larger compared to the buffer size in mobile environment. Hence, the performance of TCP Vegas is a connection that is implementing TCP Vegas. The larger the worst compared to other TCP variants. On the other hand, buffer size experience unnecessarily long delays due to large the delay estimation scheme tends to reduce to average delay. queues size in the buffer. Hence, the value of delay is larger. The RTT estimation in TCP Vegas congestion control The simulation result in the heterogeneous wired network is algorithm enables the detection of congestion in a network at tabulated in Table VIII and refers to the appendix for the early stage. Early stage detection of congestion reduces the graphical results (Fig.A3 until Fig. A10) delay in a network as the network does not have to waste time on waiting for a packet lost to conclude that congestion occurs in a network. The simulation results in the homogeneous wired-cum-wireless network is tabulated in Table IX refer to

the appendix for the graphical results.

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 95

TABLEIX network, TCP Vegas performs better than other TCP variants. SIMULATION RESULTS OF HOMOGENEOUS WIRED-CUM-WIRELESS However, the performance of TCP Vegas is the worst when it NETWORK is implemented in homogeneous wired-cum-wireless network. TCP Variants Average Average Delay Throughput (s) In both heterogeneous wired wired-cum-wireless networks, (kbps) TCP Vegas achieves unfair bandwidth allocation. On the other Vegas 809.908 0.186 hands, TCP Vegas always exhibits significant lower delay Tahoe 1873.170 0.531 compared to other TCP variants in both wired and Reno 1897.060 0.508 wired-cum-wireless network. NewReno 1921.869 0.554 SACK 1806.912 0.528 VI. FUTURE WORK This simulation experiments analyze the performance of (ii) Heterogeneous Wired-Cum-Wireless Network TCP Vegas versus other TCP variants in term of average In heterogeneous wired-cum-wireless network, TCP throughput, throughput fairness and average delay in Vegas suffers unfair bandwidth allocation when sharing the homogeneous and heterogeneous networks connection. The bottleneck link with different TCP Variants. TCP Vegas results of this simulation experiments prepared a comparison congestion control algorithm is proactive. TCP Vegas tends to medium for the performance evaluation of enhanced-TCP reduce the sending rate and its window size when incipient Vegas in IPv6 networks that will be implemented for future congestion is detects even if there is no indication of packet work. loss. As the window size is reduces, TCP Vegas is not able to For future work, we propose to enhance the performance of dominate fair bandwidth in the bottleneck link and resulted in TCP Vegas in IPv6 wireless network for both homogeneous worst throughput. Similarly to simulation experiments in and heterogeneous networks connection. We aim to modify heterogeneous wired network, TCP Vegas receive the most the sending window of TCP Vegas in order to optimize the unfair bandwidth allocation when sharing the bottleneck link throughput of TCP Vegas in wireless IPv6 network. with TCP NewReno. This is due to the reactive congestion control algorithm implemented by TCP NewReno. As long as there is no indication of packet losses, TCP NewReno APPENDIX aggressively increases its sending rate and window size, (i) Performance of TCP variants in homogeneous wired resulted in domination of the most available bandwidth of the network bottleneck link. On the other hand, TCP Vegas always has the minimum average delay than other TCP variants as TCP 2050 Vegas is able to detect congestion faster than other TCP 2000 variants. The simulation result in the heterogeneous 1950 wired-cum-wireless network is tabulated in Table X and refers 1900 to the appendix for the graphical results. 1850 TABLEX 1800 SIMULATION RESULTS OF HETEROGENEOUS WIRED-CUM-WIRELESS 1750 NETWORK

TCP Average Average Average Thrpughput (Kbps) Average 1700 Variants Throughput Throughput Delay (s) E/E E/F F/F F/E (kbps) Fairness (%) Vegas/ 168.855 / 8.443/ 0.245/ Simulation Environment Tahoe 1710.624 85.531 0.271 Vegas Tahoe Reno NewReno SACK

Vegas/ 239.918 / 11.996/ 0.190/ Reno 1617.018 80.851 0.244 Fig. A1. Performance of TCP Variants in homogeneous wired network: Vegas/ 70.457 / 3.527/ 0.261/ average throughput vs. TCP variants. (b) average delay vs TCP variants NewReno 1844.177 92.209 0.265 Vegas/ 183.918 / 9.196/ 0.231/ 0.6

SACK 1668.616 83.431 0.270 0.5 0.4 V. CONCLUSION 0.3 This simulation experiment analyze the performances of 0.2

TCP Vegas versus other TCP variants in term of average Delay (s) Average 0.1 throughput, throughput fairness and average delay in 0 homogeneous and heterogeneous wired and E/E E/F F/F F/E wired-cum-wireless networks. From our simulation results, the overall performances of TCP variants in Simulation Environment wired-cum-wireless network are poorer compared to their Vegas Tahoe Reno NewReno SACK performances in wired network. In Homogeneous wired Fig. A2 Performance of TCP Variants in homogeneous wired network: average delay vs. TCP variants

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 96

(ii) Performance of TCP variants in heterogeneous wired 100 network: 80

100 60

80 40

60 Thrpughput

Fairness(Kbps) 20 40

0 Thrpughput Thrpughput

Fairness(Kbps) 20 E/E E/F F/F F/E 0 Simulation Environment E/E E/F F/F F/E Vegas NewReno Simulation Environment Vegas Tahoe Fig. A7. Performance of Vegas and NewReno : Throughput Fairness of Vegas & NewReno versus simulation environment.

Fig. A3. Performance of Vegas and Tahoe : Throughput Fairness of Vegas

0.29

0.3

0.28 0.25 0.2 0.27 0.15 0.26

0.1 0.25 Average Delay (s) Delay Average

Average Delay (s) Delay Average 0.05 0.24 0 E/E E/F F/F F/E E/E E/F F/F F/E Simulation Environment Simulation Environment Vegas NewReno Vegas Tahoe

Fig. A4. Performance of Vegas and Tahoe : Average delay of Vegas & Tahoe Fig. A8. Performance of Vegas and NewReno : Average delay of Vegas & versus simulation environment NewReno versus simulation environment

100 100

80 80 60 60

40 40 Thrpughput Thrpughput

20 Fairness(Kbps) 20 0 0 E/E E/F F/F F/E

Thrpughput Fairness(Kbps) Thrpughput E/E E/F F/F F/E Simulation Environment Simulation Environment Vegas SACK Vegas Reno

Fig. A5. Performance of Vegas and Reno : Throughput Fairness of Vegas & Fig. A9. Performance of Vegas and SACK : (a) Throughput Fairness of Vegas Reno versus simulation environment & SACK versus simulation environment.

0.3 0.3

0.25 0.25 0.2 0.2 0.15 0.15 0.1 0.1

0.05 Average Delay (s) Delay Average 0.05 (s) Delay Average 0 0 E/E E/F F/F F/E E/E E/F F/F F/E Simulation Environment Simulation Environment Vegas SACK Vegas Reno

Fig. A10. Performance of Vegas and SACK : Average delay of Vegas & Fig. A6. Performance of Vegas and Reno : Average delay of Vegas & Reno SACK versus simulation environment versus simulation environment

(ii) Performance of TCP variants in homogeneous and heterogeneous wired-cum-wireless networks:

1111403-5656 IJECS-IJENS © June 2011 IJENS I J E N S International Journal of Electrical & Computer Sciences IJECS-IJENS Vol: 11 No: 03 97

[8] V. Jacobson, “Congestion Avoidance and Control”, In the Proceddings

100 of the SIGCOMM’88 Symposium, pp. 314-329, Aug. 1988. 80 [9] R. Dunaytsev, “TCP Performance Evaluation over Wired-cum-Wireless networks”,Phd thesis, University of Technology, 26 march 2010. 60 [10] K. Pentikousis, “TCP in Wired-Cum-Wireless Environments”, In the Proceedings of IEEE Communications Surveys & Tutorials, vol. 3, pp. 40 2-14, Fourth Quarter 2000 [11] H. Balakrishnan, V. N. Padmanabhan, “A Comparison of Mechanisms 20 for Improving TCP Performance over Wireless Links”, In the Proceedings of SIGCOMM '96 Conference proceedings on 0

Thrpughput Fairness(Kbps) Thrpughput Applications, technologies, architectures, and protocols for computer Vegas/Tahoe Vegas/ Reno Vegas/ Vegas/ SACK communications, vol.26, pp. 756-769, Oct 2006. NewReno [12] D. Katabi, M. Handley, “Congestion Control for High Bandwidth-Delay Product Networks”, In the Proceedings of SIGCOMM’02, Feb. 2007. TCP Variants [13] J. Mo, R. J. La, V. Anantharam, J. Walrand, “Analysis and Comparison Fig. A11. Performance of TCP variants in heterogeneous wired-cum-wireless of TCP Reno and Vegas”, In the Proceedings of IEEE INFOCOM 1999, network: Throughput fairness versus TCP variants. vol.3, pp.1556-1563, March 1999. [14] J. Mo, R. J. La, V. Anantharam, J. Walrand, “Analysis and Comparison of TCP Reno and Vegas”, In the Proceedings of IEEE INFOCOM 1999, vol.3, pp.1556-1563, March 1999. [15] C.Parsa, J.J Garcia-Luna-Aceves, “Improving TCP Congestion Control 0.3

over Internets with Heterogeneous Transmission Media”, In the 0.25 Proceeding of Seventh International Conference on Network Protocols, 0.2 1999. (ICNP '99). , pp. 213-221, Oct-Nov 1999. [16] W. Stevens, “TCP Slow Start, Congestion Avoidance, Fast Retransmit, 0.15 and Recovery Algorithms”, Request for Comment 2001, Internet 0.1 Engineering Task Force , Jan.1997. [17] S. Floyd and T.Henderson, “Congestion Control Techniques”, Request

Average Delay (s) Delay Average 0.05 for Comment 2914, Internet Engineering Task Force, Sep 2000. 0 [18] Y. C. Chan, C. L. Lin, “Quick Vegas: Improving Performance of TCP Vegas/Tahoe Vegas/ Reno Vegas/ Vegas/ SACK Vegas for High Bandwidth Delay Products Networks”, In the NewReno Proceeding of IEICE Trans. Communication, vol. E91B, no.4, pp. TCP Variants 987-997, Apr 2008. [19] N. Parvez, E. Hossain, “Improving TCP Performance in Wired-Wireless Fig. A12. Performance of TCP variants in heterogeneous wired-cum-wireless Networks By Using a Novel Adaptive Bandwidth Estimation network: Average delay versus TCP variants Mechanism”, In the Proceeding of IEEE Global Telecommunications Conference, GLOBECOM '04, vol. 5, pp. 2760-2764, January 2005. [20] L. A. Grieco, S. Mascolo, “Performance Evaluation and Comparison of Westwood+, New Reno, and Vegas TCP Congestion Control”, In the ACKNOWLEDGMENT Proceeding of ACM SIGCOMM Computer Communication Review, vol. We would like to thank UniMAP for provided useful 34 No. 2, pp-, April 2004. literature materials and grant FRGS 1/2010 (9003-00216) for [21] V. Jacobson, “Congestion Avoidance and Control”, In the Proceedings the financial support of this research. of ACM SIGCOMM’88, vol. 21, no. 4, pp. 314-329, August 1988. [22] J. A. Ahn, P. B. Danzig, Z. Liu and L. Yan, “Evaluation of TCP Vegas: Emulation and Experiment”, In the Proceeding of ACM SIGCOMM, vol. 25, no. 4, pp. 185-195, October 1995. REFERENCES [1] Y. Li, R. Zhang, Z. J. Liu and Q.L. Liu, “An Improved TCP Congestion First B. S. Yew is a PhD student of University Malaysia Perlis who is Control Algorithm Over Mixed Wired/Wireless Networks”, In the currently working on the research of enhancing the performance of TCP Proceeding of 2nd IEEE International on Conference Broadband Vegas in wireless IPv6 network. Network & Multimedia Technology, 2009. IC-BNMT '09., pp. 786-790, Her field of research involves evaluating the performance of current TCP Oct 2009. Vegas based on simulation via Network Simulator 2 (ns2). Wireless IPv6 is [2] M.Allman, V.Paxson and W.Stevens, “TCP Congestion Control”, implemented in ns2 via MobiWan and FHMIP extension. Request for Comment 2581, Internet Engineering Task Force, April 1999. [3] S. Biaz, N.H. Vaidya, “Distinguishing Congestion Losses from Wireless Transmission Losses: A Negative Result”, In the Proceedings of the 7th International Conference on Computer Communications and Networks,1998, pp.711-731,Oct 1998. [4] L. S. Brakmo, L. L. Peterson, “TCP Vegas: End to End Congestion Avoidance on a Global Internet”, In the Proceedings of IEEE Journal On Selected Areas In Communications, vol. 13, no. 8, pp. 1465- 1480, Oct. 1999. [5] M. Othman,M. Z. Oo, “Analysis of TCP Reno and TCP Vegas over AOMDV Routing Protocol for mobile Ad Hoc Network”, In the Proceedings of the 12th International Conference on Advanced Communication Technology (ICACT)2010, vol. 2, pp.1104-1108, Feb 2010. [6] A. D. Vendictis, A. Baiocchi, M. Bonacci, “Analysis and Enhancement of TCP Congestion Control in a mixed TCP Vegas and TCP Reno Network Scenario”, Performance Evaluation 53, pp. 225-253, 2003. [7] C. M. Tsang, K. C. Chang, “A Simulation Study on the Throughput Fairness of TCP Vegas”, In the Proceedings of the 9th IEEE International Conference on Networks, vol. 21, pp.469-474 , Oct. 2001.