Analysis of TCP under Wireless Circumstances A Performance Evaluation

Transcription

1 th International Conference on Frontiers of Information Technology Analysis of TCP under Wireless Circumstances A Performance Evaluation Rao Naveed Bin Rais 1*, Muhammad Musaddiq 2*, Mayyda Mukhtar 3*, Anees Shafiq 4*, Hafiz Muhammad Imran 5*, Muhammad Najam-ul-Islam 6 *COMSATS Institute of Information and Technology (CIIT), Lahore, Pakistan Bahria University, Islamabad, Pakistan naveedbinrais@ciitlahore.edu.pk 1, muhammad_musaddiq38@yahoo.com 2, mayydamukhtar@hotmail.com 3, aneesshafiq@yahoo.com 4, hafizimranaslam@yahoo.com 5, najam.ul.islam@gmail.com 6 Abstract With the emergence of wireless technology, and increase of mobile computing devices, the demand for the better network connectivity is the major interest of the users. Transmission Control Protocol (TCP) is the standard network protocol for communication in the Internet, but its performance drastically degrades over wireless networks, as TCP takes packet drop or collision as a sign of congestion. In this paper, we analyze TCP s performance over wireless networks using NS-2 Simulator. Unlike traditional performance evaluation methods for TCP over wireless networks, we consider different versions of TCP over different routing protocols. Moreover, we also consider the effectiveness of TCP both over infrastructure-based and infrastructure-less wireless networks. Simulation results show that TCP s performance varies from scenario to scenario both in infrastructure-based and ad-hoc wireless networks. Keywords-TCP; NS-2; Wireless Networks; Congestion; Snoop. I. INTRODUCTION With the emergence of wireless technology in the last decade, the use of handheld devices and laptops has become very common. This also brings the willingness to remain connected anytime, anywhere even when moving, most of the time using with wireless connection capabilities. The Internet is a global system of interconnected computer networks that uses the standard Internet protocol suite (TCP/IP) to serve billions of users worldwide. The Internet protocol suite is the set of communication protocols used for the Internet and other similar networks. It is commonly known as TCP/IP, because of its most important protocol, the Transmission Control Protocol (TCP). TCP is the integral part of the Internet communication, because the Internet carries a vast range of information resources and services based on TCP, such as the inter-linked hypertext documents of the World Wide Web (WWW) and the infrastructure to support electronic mail. TCP works fine over traditional wired links as presented in Figure 1(a) but its performance suffers on wireless medium [1] as shown in figure 1(b). A number of issues arise when TCP is deployed over wireless networks. TCP assumes that in wired networks a packet loss is due to congestion, but in case of wireless networks, congestion may not be the only reason for a packet loss. It may be due to many other factors including weather conditions, bit error rate, disconnection, hand-off etc. Consequently, a lot of work has already been done to improve the performance of TCP over wireless networks such as split connections like Mobile TCP (M-TCP) [2], Explicit Loss Notification (ELN) [3], Performance Enhancing Proxies [4], Indirect TCP (I-TCP) [5], etc. Despite all the efforts put in to resolve this critical problem, improving the performance of TCP over wireless networks remain an important issue, especially with the emergence of different types of wireless networks in the past decade. The performance evaluation done, to date, to verify the effectiveness of the proposed schemes usually focus on one of the two versions of TCP only, and a thorough performance analysis involving a comparison of all mainstream versions of TCP, especially on major ad-hoc wireless protocols is missing. This paper is an effort towards this direction. In this paper, we present the performance evaluation of TCP versions like Tahoe [6], Reno [7], New Reno [7], and Sack [8] over a number of different routing protocols such as Ad-hoc On-Demand Distance Vector (AODV) [9], Destination Sequenced Distance- Vector (DSDV) [10], Dynamic Source Routing (DSR) [11] and Optimized Link State Routing (OLSR) [12]. We also proposed mechanisms to improve the performance of TCP over wireless networks by using a well known protocol, Snoop [13]. We have analyzed the results that performance of TCP varies from scenario to scenario in wireless networks and in infrastructure networks handoff is the reason of inefficiency of TCP. The remainder of the paper is organized as follow. We describe the related work in section II whereas in section III, we present our performance evaluation methodology in detail. The section IV consists of the experimental results and the last section presents the conclusion. II. RELATED WORK The TCP s performance over wireless networks has been evaluated for several years. The inefficiency of TCP over wireless networks was first identified in [14], [15] in which the researchers demonstrated that the introduction of wireless links in the network degrades the performance of TCP. This is mainly due to TCP s inability to distinguish the actual reason /12 $ IEEE DOI /FIT

2 of packet loss. Specifically, a packet loss in a wired network is generally taken as a sign of congestion in the network and TCP invokes its congestion control algorithms such as Slow-Start, Fast Retransmit, which control the flow of data accordingly. On the other hand, when a packet is lost in a wireless network, it may be due to many reasons other than congestion. Regular TCP, when running over a wireless network does not distinguish between different reasons of packet loss and invokes the congestion control techniques, which result in the degradation of performance unnecessarily [16]. Several solutions have been proposed by the researchers to adapt TCP in wireless environments. Notable examples include the Snoop protocol [13], I-TCP [5], and M-TCP [2]. (a) Wired Network (b) Wireless Network Figure 1 Throughput of wired and wireless network The main idea of the Snoop protocol [13] is to cache unacknowledged data at the access point and perform local retransmissions on policies dealing with timeouts. When a duplicate acknowledgement is received for a packet, an immediate retransmission is made without waiting for the timeout. On the other hand, I-TCP [5] connection is used when a mobile host wants to communicate with a fixed host. It sends request to current base station (access point)., and the connection is established between them. The I-TCP software has two components, one on the mobile host and other on the access point. The component on the mobile host consists of special library calls. This library calls does the communication with the base station, clear to the mobile host. The second component consists of a user level UNIX process flowing data from one part of the connection into the other. Moreover, M- TCP protocol [2] is a split connection protocol where the connection between fixed and mobile host is separated at supervisor host. The TCP client at the mobile host is called M- TCP, while at supervisor host it is called SH-TCP. When the sender sends data to SH-TCP, it forwards to M-TCP which then sends acknowledgment. The SH-TCP passes this acknowledgement to the sender. When SH-TCP knows that mobile host is disconnected, it then forwards the ACK of the last byte to the sender. This forces TCP to set the window size to zero. Therefore, TCP will neither suffer from retransmit timeouts, nor it will close its congestion window. When the mobile host reconnects, the TCP sender is ready to transmit. Radio Link Layer protocol (RLP) [17] is a link layer protocol. In RLP packet is retransmitted if the sender is sure that the packet is lost or it was not received. In this way the receiver does not receive copy of any packet, which increases the efficiency of RLP. Another performance evaluation technique is AIRMAIL [17] It is an asymmetric protocol, meaning that the base station is responsible for making decisions about transmitting or receiving. This reduces the load at the mobile host. This is done by combination of automatic repeat request (ARQ) and forward error correction (FEC). Performance evaluation of TCP over wireless circumstances has also been done using different TCP versions on mobile ad-hoc network routing protocols [18]. AODV [9] routing is a routing protocol that makes its routes on demand. DSDV [10] is a table-driven routing scheme in which each entry in the routing table contains a sequence number and it updates its table after each entry and knows all possible routes. DSR [11] is a routing protocol in which each packet contains the address of each device the packet will traverse. OLSR [12] is a proactive routing protocol for mobile ad-hoc networks. The protocol inherits the stability of a link state algorithm and has the advantage of having routes immediately available when needed due to its proactive nature. III. PERFORMANCE EVALUATION METHODOLGY In this paper, we have evaluated the performance of TCP over wireless networks by performing experiments in NS-2 Simulator. The evaluation strategy takes into account the performance of TCP both over infrastructure-based and infrastructure-less (ad-hoc) networks. To the best of our knowledge, this is the first paper, which does this kind of evaluation on different types of wireless networks. Specifically, we choose IEEE standard to perform the evaluation. First, we studied and compared TCP versions such as Tahoe, Reno, New-Reno, and Sack over different ad-hoc networks routing protocols like AODV, DSDV, DSR and OLSR. We compared the results both in fixed chain scenario and mobile ad-hoc network (MANET). We have done this experiment on both scenarios i.e., both in fixed and MANET by having different number of nodes and discussed that what is the best possible combination for which the performance of TCP improves in wireless networks. Next, we implemented the Snoop protocol [13] in the NS-2 simulator at the access points (AP) to improve the performance of TCP in infrastructure-based network. We then compared this implementation with the scenario which has no Snoop protocol on APs and evaluated the results of both the scenarios. We adopted the Snoop protocol because it is the best solution in wireless networks for improving the performance of TCP on terms of throughput [16]. The authors of the Snoop protocol evaluated the protocol only with the wired cum wireless network. In this paper, we analyze the performance of this protocol with mobile nodes in the infrastructure-based network. A. Measurement Parameters The performance metrics that we chose to study in this paper are congestion window, throughput and goodput. 372

3 Throughput is defined as the ratio of the total data received by the receiver to the connection time, whereas goodput is actual data received by the receiver per unit second. We do not consider packet headers, checksum, and flags in the computation of goodput. B. Simulation Tool The results in this study are based on the simulation using the network simulator (NS-2) from the Lawrence Berkeley National Laboratory. IV. SIMULATION RESULTS We evaluated the performance of TCP using a number of experiments in NS-2 ranging from ad-hoc network with static nodes to infrastructure-based networks with mobile nodes. In the following subsections, we provide the results of some of the important experiments. A. Scenario 1: Chain Topology In this scenario, we used 15 nodes with wireless links, which are static in an ad-hoc network. We extended the work done by Foezahmedi et al. [18] who demonstrated that New Reno-AODV is the best combination for the chain topology in terms of throughput. But they only did their experiments over a network comprising of 10 nodes only. We enhanced the network from 10 to 15 wireless nodes for the detail analysis of their research. After the simulation results, we concluded that the New Reno-AODV is not the only best combination for chain topology network. After increasing the number of nodes, the simulation results are changed. One source sends the data to one sink and the network is connected in a chained manner, as shown in Figure 2, after passing through the number of nodes by using Ad-hoc On-demand Distance Vector (AODV) routing protocol, Dynamic Source Routing (DSR), Optimized Link State Routing protocol (OLSR) and table driven protocol Destination-Sequenced Distance-Vector Routing (DSDV) on the same parameters as shown in [18]. The Packet size is 1000 bytes, queue length is 100, and data rate is 2Mbps. The simulation parameters are summarized in Table 1. Table 1 Simulation Parameters for Scenario 1 Size of Network 15 nodes TCP Packet Size 1000 bytes Data Rate 2Mbps Queue Length 100 Figure 2 Chain Topology Next, we extended the network to 20 and 25 nodes. We used different versions of TCP on different routing protocols. The Reno-OLSR is the best combination for both 10 and 15 nodes networks, while New Reno-OLSR is the best combination for 25 nodes network. We observed that results vary from network to network as shown in Figures 3 and 4. The OLSR, a proactive link state routing protocol is an IP routing protocol optimized for ad-hoc network. Being a proactive routing protocol, it knows the routes to all destinations within a network and it will maintain before using them. This has been the major reason that OLSR is better than all the other routing protocols. So, as we increase the size of the network, New Reno-AODV does not remain the best combination. Figure 3 Throughput with different number of nodes using AODV and DSDV. Figure 4 Throughput with different number of nodes using DSR and OLSR B. Scenario 2: Mobile hosts Ad-hoc Network In this scenario there are 10 nodes with wireless links. Random way point model is used for mobility of the nodes. All the nodes are moving randomly at an average speed of 15 m/sec, and the average pause time is 1.5 sec. Node 0 is the sender, while node 4 is the sink. Source 1 sends data to the sink 1 by using different routing protocols on different TCP versions. Again, the aim is to identify the better combination for mobile ad-hoc wireless networks. To do a detailed analysis, we enhanced the topology used in [18], analyze the results and also used some of the protocols which are not mentioned in [18]. By using the same parameters, we made our topology. 373

4 The Packet size 1000 bytes, queue length is 100, data rate is 2Mbps. When AODV routing protocol is used on TCP Tahoe, we observed that initially the data follows the route through the wireless nodes When the nodes start moving, the route becomes changed because AODV routing protocol used the shortest path, so the route becomes By using DSDV routing scheme the data is sent late by the sender due to process of maintaining routing table. In DSR protocol the sender sends data through the route , but when all the nodes starts moving, the source 1 sends very little data through the route and when the node 5 comes in the range, it starts sending data through the route OLSR protocol shows the best performance among all the other protocols, because the routes available within the standard routing table can be useful. There is no route discovery delay associated with finding a new route. After using different versions of TCP on different protocols, we identified that Tahoe-OLSR is the combination which exhibits better performance in this scenario as shown in Figure 5. The simulation parameters are summarized in Table 2. Table 2 Simulation Parameters for Scenario 2 Size of Network 10 nodes Mobility Model Random Waypoint Speed 15 m/sec Pause Time 1.5 sec TCP Packet Size 1000 Bytes Data Rate 2Mbps Queue Length 100 Figure 5 Throughput of Ad-hoc Network with different protocols C. Scenario 3: Mobile hosts without Snoop Protocol In this scenario, we used TCP Reno in infrastructure-based network. There are 12 nodes with wireless links, 4 APs while these APs are connected through Ethernet. There are two senders and two receivers. Data rate of both the senders is 8Mbps, interval is sec and the packet size is 1024 bytes. Hierarchical routing is used in this network. There is no Snoop protocol at APs. One of the sources is moving at a speed of 10m/s while sending data, and the other is stationary. The queue size is maintained at 10. Both sources are in different domains. The mobile source is in domain 2 and it sends data to the sink which is in domain 4. While the stationary source is in domain 3 and it sends data to the sink in domain 2. The goodput of both the sources are shown in Figure 6(a). TCP Reno is used in this scenario, so whenever packet drops or congestion occurred the congestion window does not goes down to 1. It goes to half of current window as shown in Figure 6(b). In this scenario, when the mobile source gets out of its AP s range and enters into the range of an intermediate AP, the data is handed over to the new AP. Thus, there is decrease in the performance of TCP. The throughput of the mobile source decreases from t=5.0 sec to t=17.0 sec, because it is moving towards the other AP. Packet drops in the scenario are shown in Figure 6(c). The simulation parameters are shown in Table 3 for this scenario. Table 3 Simulation Parameters for Scenario 3 Size of Network 12 nodes Data Rate 8Mbps Interval sec TCP Packet Size 1024 bytes Speed 10 m/s Queue Length 10 D. Scenario 4: Mobile hosts with Snoop Protocol Next, we modified Scenario 3 to implement the Snoop protocol on APs. All other parameters are the same. In this scenario, we compared the performance of TCP with that of Scenario 3, where the Snoop protocol was not used. With the help of the Snoop protocol, we achieved higher goodput and throughput for both senders as shown in Figures 7(a) and 7(c) respectively. In Figure 6(c), the throughput of the mobile source at t=25 sec is 140kbps and the throughput of stationary source at t=25 sec is 150kbps. On the other hand, in Figure 7(c) the throughput of the mobile source at t=25 sec is 190kbps, and that of the stationary source is 160kbps. The congestion window shown in Figure 7(b) is much smoother than the congestion window presented in Figure 6(b). It is due to the fact that there are less packet drops when we used the Snoop protocol at the APs. The simulation shows that by using the Snoop protocol, we can improve performance of TCP over infrastructure-based wireless networks, even with mobile nodes. V. CONCLUSIONS One of the most challenging and interesting recent trends in computer networks is the integration of wireless communications. With the increasing importance of wireless communication, and the popularity of TCP on fixed networks, we are in need of more reliability of TCP in wireless networks. This improvement in TCP can be address by keeping the problems of wireless connections in the mind. In this paper, we illustrated, using simulation results, the problems caused in the wireless networks and the difficulty arises due to mobility in the wireless network. We demonstrated their negative effects on the performance of TCP. 374

5 (a) Goodput (a) Goodput (b) Congestion Window (b) Congestion Window (c) Throughput Figure 6 Infrastructure Network without Snoop (c) Throughput Figure 7 Infrastructure Network with Snoop 375

6 We deduced the following three conclusions from this study: 1. TCP performance varies from scenario to scenario in wired and wireless networks. 2. The performance of TCP decreases by increasing the number of wireless nodes in the network and vice versa. 3. The efficiency of TCP improved by using the Snoop protocol in infrastructure-based network. REFERENCES [1] H. Balakrishnan, V. Padmanabhan, S. Seshan, and R. Katz, A comparison of mechanisms for improving TCP performance over wireless links, In Proceedings of ACM SIGCOMM 96, Stanford University, California, August, [2] I. Rhee, N. Balaguru, and G.Rouskas, M-TCP: Scalable TCPlike Congestion Control for Reliable Multicast, In Proceeding of IEEE INFOCOM, pp , March [3] S. Floyd, TCP and Explicit Congestion Notification, Computer Communication Review, vol. 24, No. 5:10-23, [4] J. Border, M. Kojo, J. Griner, G. Montenegro, and Z. shelby, Performance Enhancing Proxies, IETF Internet draft, June [5] A. Bakre and B. Badrinath, I-TCP: Indirect TCP for Mobile Host, In proceeding of 15 th International Conference on Distributed Computing Systems, May [6] V. Jacobson, Congestion Avoidance and Control, In Proceeding of SIGCOMM, pp , [7] Hala Elaarag, Improving TCP Performance Over Mobile Networks, ACM Computing, vol. 34, No. 3, pp , September [8] M. Mathis, J. Mandavi, S. Floyd, and A. Romanov, TCP Acknowledgment Options, IETF RFC 2018, [9] Perkins, C., Elizabeth, M., R., and Samir, R. D., (1999), Ad hoc on demand distance vector (AODV) routing, Internet draft, IETF, January [10] Perkins, C., E., and Bhagwat, P., (1994), Highly dynamic Destination-Sequenced Distance-Vector routing (DSDV) for mobile computers, Proc. of the DIGCOMM 94 Conference on Communications Architectures, Protocols and Applications, pp , August [11] Johnson, D. and Maltz, D. (1996), "Dynamic Source Routing In Ad hoc Wireless Networks," In Mobile Computing (ed. T. Imielinski and H. Korth), Kluwer Academic Publishers, Dordrecht, The Netherlands. [12] T. Clausen, P. Jacquet, A Laoiti, P Minet, P. Muhlethaler, A Qayyum, L Viennot, Optimized Link State Routing Protocol, IETF RFC 3626, October [13] Hari. Balakrishnan, S. Seshan, Elan Amir, and R. Katz, Improving TCP/IP performance over wireless Networks, In Proceedings of ACM MOBICOM 95, University of California, [14] Mun Choon Chan and Ramachandran RAmjee, Improving TCP/IP Performance over Third Generation Wireless Networks, IEEE INFOCOM, [15] Wang Long, and Wan Zhenkai, Performance Analysis of Improved TCP over Wireless Networks, Computer Modeling and Simulation ICCMS '10. Second International Conference, pp , 22-24, January [16] Sarma Vangala, and Miguel A. Labrador, Performance of TCP over Wireless Networks with the Snoop Protocol, In Proceedings of the 27th Annual IEEE Conference on Local Computer Networks, [17] Hala Elaarag, Improving TCP performance over mobile networks, ACM Computing, vol. 34, No. 3, pp , September [18] Foezahmedi, Sateesh Kumar Pradhanu, Nayeema Islam, and Sumon Kumar Denath, Performance Evaluation of TCP over Mobile Ad-hoc Networks, (IJCSIS) International Journal of Computer Science and Information Security, vol. 7, No 1,