Analysis of DelAck based TCP-NewReno with varying window size over Mobile Ad Hoc Networks
Vol. 10 No. 1 January 2012 International Journal of Computer Science and Information Security Publication January 2012, Volume 10 No. 1 . Copyright � IJCSIS. This is an open access journal distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 10, No. 1, January 2012 Analysis of DelAck based TCP-NewReno with varying window size over Mobile Ad Hoc Networks Parul Puri1 Gaurav Kumar2 Bhavna Tripathi3 Dr Gurjit Kaur4 Department of Electronics & Communication Engineering Assistant Professor, Jaypee Institute of Information Technology, Department of Electronics & Communication Engineering Noida, India. School of ICT, email@example.com Gautam Buddha University, firstname.lastname@example.org Greator Noida, India. email@example.com firstname.lastname@example.org Two key requirements of any network are reliable data transfer and congestion control. The transmission control Abstract—In this paper, we study TCP performance over multi- protocol (TCP) was designed to provide reliable end-to-end hop wireless networks that use IEEE 802.11 protocol for access. delivery of data packet in the wired networks. However, For such networks NewReno is the most deployed TCP variant unlike wired networks wireless networks suffer from many that handles multiple packet losses efficiently. It is shown that problems, such as packet losses due to congestion, node the delayed ACK scheme substantially increases the TCP mobility, high bit errors, medium access contention due to throughput. We propose an approach to improve the hidden terminals, and so on. Hence, in order to apply TCP in a performance of half-duplex and asymmetric multi hop networks wireless environment, TCP needs some modifications. widely employed for mobile communication. Our approach is based on optimizing the timer duration of the delayed ACK Further, keeping in mind the basic characteristic of a TCP scheme and varying the window size. Simulations have been scheme the acknowledgement (ACK) packets need to be carried on NS2 for TCP-NewReno variant using DSDV and transmitted from TCP sink to TCP source, against the flow of AODV routing protocols. TCP data packets. This results in simultaneous arrival of TCP data and ACK packets which can cause collisions and even Keywords: Multi-hop wireless networks, TCP, Newreno, DelAck, packet losses [2, 3]. As a result, there is a huge degradation in DSDV, AODV. throughput in multi-hop networks . I. INTRODUCTION At the MAC level, each data packet transmission is a part In the last few years, many research works have focused of four-way handshake protocol, which is intended to reduce on multi-hop wireless networks, in which relaying nodes are the collision probability. The handshake reduces the in general mobile, and communication needs are primarily probability of hidden-terminal collisions, but it does not between nodes within the same network. In such networks, a eliminate them. This limits the number of packets that can be number of intermediate nodes whose function is to relay transmitted simultaneously in a wireless network without information from one point to another point carry out collisions. The main factor affecting the TCP performance in communication between the two end nodes. The application multi-hop wireless networks is the contention and collision can be useful in various fields, especially because it uses between ACK and data packets caused by taking the same wireless means of communication, hence saving the hassle of path. Thus, in order to improve the TCP throughput, we shall laying down wires in already crowded or remote terrains. try to decrease the ACK flows by using the delayed ACK People working in collaboration and places in remote scheme, where an ACK is transmitted for every d packets, locations can connect through it. Activities which require defined by the DelAck number, that reach the destination . working at locations having no ground infrastructure, like However, to avoid a deadlock, and if d packets do not arrive, patrolling, disaster hit areas and rural areas, can be carried out an acknowledgement is generated after some time interval using this technology. Some important applications are also without further waiting. being developed on the basis of this technology which can be The throughput of a network is limited by two windows: used by armed forces in rescue and war time scenarios . the congestion window and the receive window. The TCP sender uses a congestion window (cwnd) in regulating its 62 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 10, No. 1, January 2012 transmission rate based on the feedback it gets from the RFC 5681 mandates that an acknowledgement be sent for at network . Whereas, the receive window size sets a limit on least every other full-size segment, and that no more than the amount of data that can be sent unacknowledged. Earlier 500ms expire before any segment is acknowledged. researches on TCP performance over multi-hop wireless Basically, the delayed acknowledgement procedure defines networks  have shown that for static chain topology it is beneficial to limit the maximum receive window size of TCP two terms: DelAck number and Time interval. The DelAck number d defines the number of packets for which the receiver sink to around n/4, where n is the number of nodes; and any further increase in the maximum window size causes more waits before sending an acknowledgement. By using delayed acknowledgement mechanism the numbers of collisions and deterioration in the throughput. However, the issue of limit on an optimum window size for mobile topology acknowledgments required are reduced. As acknowledgments are also parts of traffic, the load over channel decreases. Thus, is left unaddressed. using this concept the throughput is increased. But this is not It is also seen, for a fixed small size of maximum window always the case; there are some situations where delayed size, the delayed ACK does not outperform the standard TCP acknowledgment leads to reduction in bandwidth. Studies version since most of the time, the window size limits the have shown d = 2 gives an optimum performance. number of packets that can be transmitted by the sender to less Second parameter of the delayed acknowledgement than d. So, the delayed ACK scheme has to wait for the timer to expire before generating an ACK; and the sender cannot procedure is the Time Interval (Fig. 1). A timer is set by the TCP, depending on which DelAck procedure is modified. transmit packets during that time. Hence, the time interval plays a critical part of TCP system with DelAck scheme. Now the acknowledgement is sent when the two packets are received or if the timer goes off, whichever occurs first. Tahiliani et al in  has studied the performance of TCP variants such as Tahoe, Reno, NewReno, Sack, and Vegas We aim to study the effect of the delayed acknowledgement procedure on TCP throughput over multihop wireless links. over various routing protocol. They have analyzed that TCP NewReno and Sack perform better in comparison to the other Jiwei Chen et al  has studied that increasing the value of schemes. In this paper, the NewReno variant of TCP is tested DelAck number does not always show a positive increase in as it is the most deployed one. We propose an approach to the throughput. In some situations it has proved to be improve the TCP performance by simulating the delayed ACK deteriorating also. This is so because if a large DelAck scheme with an optimum time interval and by varying the number is chosen it will cause a large burst of packets to pass receive window size for the same size of congestion window thereby increasing interference. Keeping in view this adverse (cwnd) for mobile topology. We choose one proactive routing affect we have kept our DelAck number to be 2 and focus our protocols: Destination Sequenced Distance Vector (DSDV) as study on the Time interval aspect. well as one reactive routing protocols: Ad hoc On demand Distance Vector (AODV) for our study since they are accepted as the standard routing protocols for multi-hop wireless networks . II. Related Work In Reference , G. Holland et al uses a new metric called expected throughput to compare the performance by measuring the differences in throughput with varying number Figure 1. Role of DelAck and Time Interval in TCP communicattion of hops. Further the authors have studied the effects of mobility on TCP Reno‟s performance in mobile ad hoc IV. Window Size networks. This metric will be used in our paper and will be discussed in detail in Section V. In order to limit the impact of congestion, TCP uses a special kind of buffer called Sliding (Receive) Window. Ammar Mohammed AI-Jubari  has shown that the Receive window size indicates the buffer size of the receiver. delayed acknowledgment strategy can improve TCP In other words, window size is the maximum number of throughput up to 233% compared to the regular TCP over packets (bytes) a source can transmit before receiving an multi-hop wireless networks. acknowledgement from the receiver. By controlling the Jiwei Chen  has tried to explain the effect of receive window size, a receiver can control the rate at which other window size on the TCP throughput, but have restricted the hosts send data to it. For the small window size, the number of research to static topology only. packets transmitted to the receiver is less. But the number of acknowledgements transmitted in this case will be comparatively larger and will cause collision with data III. Delayed ACK Scheme packets, thus reducing the throughput. On the other hand, if RFC 831 first suggested a delayed acknowledgement the window size is too large, number of acknowledgements (DelACK) strategy, where a receiver doesn't always decrease. However, as the receiver buffer size is more, number immediately acknowledge segments as it receives them. This of packets transmitted by the sender host increases thereby recommendation was carried forth and specified in more detail causing bursty traffic. This causes interference and packet in RFC 1122 and RFC 5681 (formerly known as RFC 2581). losses depending upon the path length. Thus, there exists an 63 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 10, No. 1, January 2012 optimum window size for which the channel gives maximum scene file. So, we calculate the expected throughput using (1) throughput. We aim to find the size of this optimum size of as follows: the Sliding window. t i Ti (1) V. Simulation Setup and Methodology E xp ected throughput i 1 Simulations have been done on ns-2 , a discrete event ti 1 i simulator. The simulations were carried for multihop wireless Practical Throughput is obtained from the simulations. Both static and mobile topologies. expected and practical throughputs are then compared in terms A. Multihop Wireless Static Topologies of the percentage achieved of the expected throughput calculated as follows: A linear string topology of 8 nodes was designed, similar Practical Throughput (2) to the one used in . A single TCP connection with variable Percentage Achieved = % number of hops (1-7) was studied. The nodes were configured Expected Throughput to use 802.11 MAC protocol with the following parameters. VI. RESULTS AND ANALYSIS Distance between two nodes was 250 metres. This distance is same as the maximum transmission range. Radio propagation A. Multihop Wireless Static Topologies model used was Two-ray ground reflection model. The Tables I and II show the throughput (in Kbps) obtained for channel data rate was 2 Mbps, TCP packet size was 1460 each variant of TCP with DSDV and AODV routing protocols bytes and the maximum window size was 32. With the above respectively. These results will be used for calculating the mentioned parameters fixed and varying the TCP protocol, expected throughput values as explained in Section V. routing protocol and TCP sink results were taken. The results have been discussed in Section VI. Our studies show that NewReno variant of TCP gives the most optimum performance as compared to other variants for B. Multihop Wireless Mobile Topologies both the routing protocols. This is because of the fact that Our network model constitutes of 25 nodes in a 1500 x NewReno is more capable in handling multiple packet losses 400 m2 flat, rectangular area. Movement of nodes was from a single window of data as compared to other TCP according to the mobility patterns generated by the mobility variants. Hence, for mobile topologies we carry out our pattern generator offered by ns-2; which is based on random analysis for the NewReno TCP scheme. waypoint mobility model. In this model, each node picks a random destination. Once it arrives to the destination it pauses As is known, the performance of TCP depends on the for some time and then picks another destination. This routing protocols as every routing protocol has a different procedure is followed throughout. The mean speed of the technique to handle link failures and to form routes. From our nodes was taken 10m/s and the pause time was 0 sec. The results, it can be seen in static topologies performance of simulation results are based on an average throughput of 25 proactive routing protocol (DSDV) is better in terms of mobility patterns. The parameters were same as those taken throughput as compared to reactive routing protocol (AODV). for static topologies. Here, the TCP-NewReno variant was The reason is that proactive protocols maintain a routing table. studied with variations in TCP sink, routing protocol and However, in reactive protocols route calculation is on-demand window size. Simulation results are discussed in the Section basis which causes some delay in sending data. Also, DSDV VI. has lesser number of control packets which decreases the C. Performance Metric number of collisions. Throughput has been used as the performance metric. Further, an improvement in throughput is observed when Throughput was measured for fixed sender and receiver nodes DelAck is used for all TCP variants over DSDV and AODV over the entire period of the connection. TCP cannot routing protocols. determine the cause of packet loss, and considers congestion the reason behind the losses. Thus, the throughput so obtained B. Multihop Wireless Mobile Topologies is always less than the optimal value. In order to compare the Tables III and IV show the throughput (in Kbps) obtained difference, we use another metric called the expected for the NewReno variant of TCP with DSDV and AODV throughput. Expected throughput gives an upper bound on the routing protocols respectively. Throughput values have been TCP throughput. Expected throughput is calculated using the obtained by varying the characteristics of TCP sink such as throughput values obtained in the static topologies. If t i = time, window size and delay interval. Ti = throughput, where i = hops (ranges from 1 to 7). Hence t1 means "amount of time source and destination were 1 hop far Based on the simulation results Fig. 2 to Fig. 7 have been from each other". Similar explanation comes for throughput. plotted and will be further analyzed. T2 means "throughput when source and destination were 2 hops far from each other". The values of Ti are those obtained from simulating static topologies and ti is obtained from the 64 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 10, No. 1, January 2012 TABLE I. THROUGHPUT (IN KBPS) USING DSDV Tahoe Reno New Reno Sack No of Hops Without With Without With Without With Without With DelAck DelAck DelAck DelAck DelAck DelAck DelAck DelAck 1 752.19 802.40 752.19 802.40 752.19 802.39 752.19 802.39 2 376.60 402.15 376.60 402.15 376.60 402.15 376.60 402.15 3 251.15 271.74 224.98 271.74 224.98 271.74 165.08 271.74 4 173.44 185.36 164.70 180.06 160.00 185.58 179.79 184.50 5 152.62 164.44 140.10 159.88 155.98 121.59 154.48 160.52 6 141.22 148.07 124.32 143.43 143.05 152.84 144.65 151.25 7 133.16 139.06 123.75 131.73 135.36 148.58 74.26 79.77 TABLE II. THROUGHPUT (IN KBPS) USING AODV Tahoe Reno New Reno Sack No of Hops Without With Without With Without With Without With DelAck DelAck DelAck DelAck DelAck DelAck DelAck DelAck 1 757.76 805.10 757.76 805.10 757.76 805.10 757.76 805.10 2 379.15 403.50 379.15 403.50 379.15 403.50 379.15 403.50 3 198.02 222.21 199.60 222.21 211.56 222.21 203.98 217.61 4 151.24 178.50 127.64 154.55 152.46 177.88 150.65 174.58 5 127.37 152.30 113.98 137.77 130.05 152.47 126.17 150.17 6 116.77 136.02 105.44 125.04 119.80 135.58 118.47 133.21 7 51.81 75.06 53.89 99.08 56.25 71.39 42.29 107.01 TABLE III. THROUGHPUT (IN KBPS) USING DSDV Without DelAck- DelAck- DelAck- Window Size DelAck 100 ms 120 ms 140 ms 2 496.36 520.60 535.20 523.08 4 509.72 531.44 535.68 532.44 6 491.44 524.64 528.64 538.68 8 525.77 564.11 546.54 559.28 20 519.84 560.02 544.97 547.22 32 524.11 560.76 570.00 549.33 Expected Thpt 592.48 634.31 634.31 634.31 TABLE IV. THROUGHPUT (IN KBPS) USING AODV Without DelAck- DelAck- DelAck- Window Size DelAck 100 ms 120 ms 140 ms 2 513.96 539.28 525.88 538.76 Figure 2.Throughput (in Kbps) using DSDV with varying 4 498.84 549.08 544.08 534.04 Window Size 6 504.60 555.96 540.68 546.08 For window sizes 2, 4, and 6 the throughput is lesser than 8 501.68 554.44 543.28 542.80 the optimum window size - 8 for different delay intervals. 20 514.80 559.24 537.44 543.28 This decrease in throughput for small window sizes at higher 32 503.16 554.44 547.52 548.12 intervals is evident as for small window sizes the buffer Expected Thpt 595.17 632.91 632.91 632.91 capacity of receiver is small. Hence, the sender can now send a limited number of packets until it has received From Fig.2 it is seen that DSDV gives a maximum acknowledgements for all packets in that window. However, throughput of 570 Kbps for window size of 32 and a delay of as the timer interval is more, receiver remains idle for a longer 120 ms. In this case 90% of the expected throughput is duration before sending the acknowledgement. This results in achieved. However, for other delay intervals (0, 100, and 140 a decrease in throughput. On similar grounds, the adverse ms) window size-8 outperforms all other window sizes effects of elevated idle time are observed for 140 ms delay including 2, 4, 6, 20, and 32. The percentage achieved is 89% interval for all window sizes. This indicates the limitation on of expected throughput for window size - 8. the value beyond which delay interval should not be increased. 65 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 10, No. 1, January 2012 53, 44, 51) Kbps are obtained for window sizes 2, 4, 6, 8, 20, and 32 respectively in comparison to gains (12, 45, 36, 42, 23, and 44) Kbps for 120 ms delay. Figure 3. Throughput (in Kbps) using AODV with varying Window Size In case of AODV, as seen from Fig. 3, peak in throughput obtained is 559 Kbps at a window size of 20 with delay 100 Figure 5. Throughput (in Kbps) using AODV with varying Window Size and Time Intervals ms. It has achieved 88% of the expected throughput. In comparison to the DSDV protocol, AODV has some variations in terms of the optimum window size and delay Fig. 6 gives a comparison of the expected throughput interval. As seen from Fig. 3, peaks in throughput values are values and the practical throughput values obtained through obtained for larger window sizes such as 20 and 32 for simulations. The practical throughput values taken for different time intervals, in comparison to DSDV where the comparison are the maximum values obtained for respective optimum size of window for different time intervals was 8. In time intervals (100, 120, and 140 ms). It is seen, in order to terms of the delay interval, Fig. 5 shows that best performance achieve practical throughput values as close to the expected for AODV is obtained for DelAck=100ms. Any further throughput, it is important to select time interval in increase in the delay interval degrades its performance. conjunction with the window size. Overall, performance of DSDV is better in comparison to the AODV protocol. Fig. 7 gives the values of the respective time intervals for different window sizes (2, 4, 6, 8, 20, and 32) which give maximum throughput. As can be seen for DSDV, window sizes 8 and 20 give maximum throughput of 564 Kbps and 560 Kbps at 100 ms time interval. For AODV all window sizes give maximum throughput at 100 ms time interval. Figure 4. Throughput (in Kbps) using DSDV with varying Window Size and Time Intervals Further, from Fig. 4 we analyze the gain in the throughput values obtained with DelAck and without DelAck. As expected theoretically, a significant amount of gain is obtained using DelAck. Also, the amount of gain is dependent on the Figure 6. Comparison of Expected and Maximum Practical Throughput (in two parameters, delay and window size. For smaller window Kbps) using DSDV and AODV with varying Window Size and Time sizes (2 and 4) the gain is more for 120 ms delay. For eg. for Intervals window size 4, throughput gain is 39 Kbps for 120 ms delay while for delays 100 and 140 ms the gain is 24 and 27 Kbps respectively. For larger window sizes (8 and 20) a delay of 100 ms gives the highest throughput. In case of AODV, Fig. 5 shows maximum gain is achieved for 100 ms delays for all window sizes. For 100 ms delay, gains as high as (25, 50, 51, 66 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 10, No. 1, January 2012 case of AODV, 100 ms delay with variable window size gives optimum throughput. Currently, we are also analyzing the effect of number of nodes on the choice of window size and time interval. Testing our approach in a real test-bed experiment, to show its efficiency in the real TCP, is a part of our future work. References  M. Gerla and J.T.-C. Tsai, “Multicluster, mobile, multimedia radio network,” ACM/Baltzer Journal of Wireless Networks, vol. 1, no. 3, pp. 255-265, 1995.  G. Holland and N. Vaidya, “Analysis of TCP performance over mobile ad hoc networks,” in Proceedings of ACM/IEEE MOBICOM, Seattle, Washington, August 1999 .  T. Kuang, F. Xiao, and C. Williamson, “Diagnosing wireless TCP Figure 7. Plot of maximum throughputs (in Kbps) obtained for different performance problems: a case study," in Proceedings of SCS SPECTS window sizes with their delay intervals using DSDV and AODV Conference, Montreal, PQ, pp. 176-185, July 2003.  M. Tahiliani, K.C. Shet, and T.G. Basavaraju, “Performance evaluation VII. Conclusions and Future Work of TCP variants over routing protocols in multi-hop wireless networks,” ICCCT‟10. Through simulation we have studied the effect of delayed  A.M. Al-Jubari and M. Othman, “A new delayed ACK strategy for acknowledgment with variations in time interval for various TCP in multi-hop wireless networks,” Information Technology receive window sizes on TCP NewReno in mobile multi-hop (ITSim), pp. 946 – 951, June 2010 . wireless networks. It is evident from the results that, there  Pasi Sarolahti, “Linux TCP,” Nokia Research Centre. exists a tradeoff between the time interval and window size.  Z. Fu, P. Zerfos, H. Luo, S. Lu, L. Zhang, and M. Gerla, “The impact We propose that maximum throughput can be achieved by of multi-hop wireless channel on TCP yhroughput and loss," in Proceedings of IEEE INFOCOM, San Francisco, CA, April 2003. selecting an optimum time interval for a particular window  E. Jiwei Chen, Yeng Zhong Lee, Mario Gerla, and M.Y. Sanadidi, size. Further, it is seen choice of window size and time “TCP with delayed ack for wireless networks,” in Broadband interval varies with the routing protocols also. Results show Communications, Networks and Systems, pp. 1-10, October 2006. for DSDV a time interval of 120 ms and a large window size  K. Fall, K. Vardhan, “The ns manual,” The VINT Project, January of 32 gives a peak in throughput. However, a window size of 8 2009. gives the most optimum results for various delay intervals. In  M. Gerla, K. Tang, R. Bagrodia, “TCP performance in wireless multihop networks,” in Proceedings of IEEE WMCSA, New Orleans, LA, February 1999. AUTHORS PROFILE Bhavna Tripathi received the B.Tech degree in Parul Puri received the B.Tech degree in Electronics Electronics & Communication Engineering from & Communication Engineering from National Institute Gautam Buddh Technical University, Lucknow , India of Technology, Hamirpur, H.P., India. She is currently in 2010 and is currently pursuing the M.Tech degree in pursuing the M.Tech degree in Electronics & Electronics & Communication Engineering from Communication Engineering from Jaypee Institute of Jaypee Institute of Information Technology, Noida, Information Technology, Noida, India. India. She has worked as a Patent Analyst in a Her current research interests include leading legal process outsourcing firm CPA Global, digital and wireless communication, digital signal Noida. She has hands on experience in patent analysis, processing, simulation of telecommunication systems and radio-navigation patent infringement, and patent portfolio management in various technology systems. domains including „Speech‟, „IP Multimedia Subsystem architecture‟, and „Biometrics‟. Dr Gurjit Kaur has been an Assistant Professor with Her current research interests include spread-spectrum the Gautam Buddha University, Greater Noida, India. communication, multi-carrier communication, channel coding, and channel She received her ME and Ph.D degrees both from the fading. PEC University of Technology, Chandigarh in 2003 and 2010 respectively. She has been a topper Gaurav Kumar received the B.Tech degree in throughout her academic career and has received the Electronics & Communication Engineering from gold medal from Honorable President of India for Kurukshetra University, Kurukshetra, India in 2008 being overall topper at Punjab Technical University, and is currently pursuing the M.Tech degree in Jalandhar. Electronics & Communication Engineering from Her professional research areas are Wireless and Optical Jaypee Institute of Information Technology, Noida, Communication. She has many research papers of national and international India. repute to her credit. She has served as a reviewer of journals and conferences. His current research interests include digital and wireless communication, resource allocation for broadband wireless transmissions, simulation of telecommunication systems and image processing in VHDL. 67 http://sites.google.com/site/ijcsis/ ISSN 1947-5500