TCP; WiLD). I. INTRODUCTION. Wireless systems have become the fastest growing segment in the telecommunications, due to the advantages over wired.
Performance Evaluation of the New Algorithm of the TCP Protocol for a Long Distance Wireless Link in the Galapagos Islands Román Lara-Cueva1, Member, IEEE, Gonzalo Olmedo2, Mishel Acosta3, and Javier Sandoval4 Grupo de Investigación en Sistemas Inteligentes (WiCOM-Energy) and Ad Hoc Networks Research Center (CIRAD) Departamento de Ingeniería Eléctrica y Electrónica, Universidad de las Fuerzas Armadas ESPE Sangolquí-Ecuador, UIO 171-5-231B Email: {1ralara, 2gfolmedo, 3amacosta, 4vjsandoval}@espe.edu.ec Abstract—In this paper we evaluate the performance of the new algorithm of the TCP protocol for a long distance wireless link in the Galapagos Islands. In order to achieve this goal, it has been started from a study of different standards and tools for this project, besides of the amendments to the TCP protocol for wireless environments. Finally, the network was implemented and preceded to inject traffic to evaluate the performance of the TCP protocol. The results showed that the modified TCP protocol with a window of 15 is more appropriate, however, it was noted the need to optimize the sensing mechanism of the error type to improve wireless TCP performance by using cross layer design. With the intention of optimizing the TCP performance, we have mathematically modeled the RTT value that should be used to calculate the appropriate RTO. Keywords—ACK Time Out; NACK; RTO; RTT; wireless TCP; WiLD)
I.
INTRODUCTION
Wireless systems have become the fastest growing segment in the telecommunications, due to the advantages over wired systems such as less time for installation, lower costs of infrastructure and maintenance, and a faster return investment. Despite all these advantages, wireless links have major difficulties in terms of reliability, time varying, fading problems, mobility, limited bandwidth and large propagation times [1]. For these reasons and because of the need of a faster data transmission emerges the necessity to modify the TCP protocol [2]. TCP is a connection-oriented protocol, which was created based on the assumption that congestion network is the primary cause of packet loss, which is true in wired networks where the bit error rate (BER) is negligible. Although TCP is able to retransmit lost packets, it interprets packet loss by connection problems as a sign of network congestion, booting the congestion avoidance procedure that abruptly reduces the congestion window; this mistaken assumption produces that applications have high values of latency and a lower bitrate in wireless networks. Among the main difficulties that the TCP protocol must overcome in wireless environments are: the undesired signals over the wireless channel, which causes transmission errors, the multipath fading, and the larger propagation times. The original TCP protocol has been modified to suit wireless environments. There have been some solutions to
adapt TCP to wireless networks as in [2], [3], and [4], which use mechanisms to solve the problems at the link layer by FEC techniques and accelerated retransmissions. At the transport layer level, the main modifications made to TCP for wireless environments have been proposed in [5], [6] and [7], which use selective ACK mechanisms. While [8], [9], [10], and [11] attempts to estimate the available bandwidth by using different techniques, and adapts the sending rate according to the network conditions. The algorithm proposed in [12] and implemented in [13] uses the idea of extending the congestion window with 15 (NACK+15) and 20 (NACK+20) additional segments, from the sender receives a negative ACK, new TCP achieves a partial improvement in terms of throughput before the congestion window is reduced. In order to increase the use of the available bandwidth, it is required to optimize the communication between the transport layer and the link layer as described in [14] [15] and [16]. Those three researches state that the round trip time (RTT) of the transport layer depends on the values determined by the link layer affecting the performance of the TCP protocol in wireless networks. Unfortunately the proposed solutions involve any kind of wireless network, and they do not take into account the characteristics of the link layer for a particular standard. Previously, we evaluate the performance of the new TCP algorithm in short distance over WiFi and WiMAX links, this algorithm shows improvements up to 200% of the throughput, meanwhile the delay was reduced up to 100% by using TCP NACK+15 [17]. The aim of this work is to evaluate the performance of the new TCP algorithm over a point-to-point long distance wireless link, between two islands in the Insular Region of the Ecuador. The results shows the need to perform a cross layer communication to determine the appropriate RTO value to increase the performance of the TCP protocol according to the mathematical model described in [14] and [18]. The rest of this paper is organized as follows: section II focuses on the methods and materials used for this project, section III develops the planning and implementation of the network, section IV shows the results obtained, while in section V the results are analyzed, and the optimization to compute the appropriate RTO is performed; finally section VI presents the conclusions and future works.
978-1-5090-2458-2/16/$31.00 ©2016 IEEE
II.
MATERIALS AND METHODS
This section details the materials and methods used to design and simulate the radio link between two islands in the Insular Region of the Ecuador. The methodology used in this project consists of three main phases: first the investigation of the standards IEEE 802.11, IEEE 802.16; the TCP protocol, the elements of hardware and software needed to design and simulate a long distance wireless link with each technology. The second part contains the installation of the equipment and the necessary tests to evaluate the performance of the TCP in wireless environments. Finally the third part performs the results analysis. For the materials, a review of the elements needed to deploy a long distance link was made for each technology (WiLD, WiMAX and proprietary equipment). The equipment selected to perform the radio link was Motorola PTP 58600. This radio equipment combines the speed and reliability of data transmission; provide MIMO, iOFDM, spatial diversity and smart dynamic frequency selection.
III.
PLANNING AND IMPLEMENTATION OF THE NETVORK
Good planning was seen as a series of stages, ranging from the generation of a set of criteria to design the network, based on the needs; a technical site survey (TSS) at each place preselected until the final selection and the implementation of the network. The TSS was needed to validate the predesign or to modify it. Based on the TSS it was found the optimal location of the sites to connect Santa Cruz Island and San Cristóbal Island. This was corroborated by the analysis of simulation software. The sites selection took into account the physical, technical and administrative facilities, as well as the simulations and TSS results. The places selected because of the best location, conditions and being the highest points in each island were: San Cristóbal in the tower at San Joaquin Hill (00º53'51.93"S, 89º31'03.30"W), and Santa Cruz at Crocker Hill (00º38'36.34"S, 90º19'55.46"W). Therefore the network consists of a transmitter and a receiver without repeaters, as depicted in Fig. 1.
4 TCP TCP 15 TCP 20
3.5 3 2.5
Bitrate
It was mainly focused on the study of both the simulation software, which helps to design the radio link, and the traffic injector, which evaluates the performance of the network. Finally, it was made a review of the virtual machines, which have implemented the new algorithm of TCP windows NACK+15 and NACK+20. In addition, we use an intrusive technique by Distributed Internet Traffic Generator (DITG) as traffic injector, which is an open code tool that follows the client–server model and allows generating traffic to analyze protocols at application, network and transport layers, a distributed performance measurement tool, able to calculate the One Way Delay (OWD), the RTT, the packet loss, the jitter and throughput was designed [18]. The tests performed consisted of the traffic injection by periods of 5 minutes with different number of packets at rates of 1, 2, 3 and 4 Mbps with each version of the TCP protocol. Finally, Simulation of Radio Electric Networks (Sirenet) was selected to simulate the radio link. This software facilitates the planning and management of networks; it works with all kinds of radio services and technologies using advanced computational algorithms [19]. Besides Sirenet, Motorola PTP LINK Planner was also used. This software seeks to eliminate as many assumptions in the design process of the link, and it is designed to point-to-point wireless Ethernet links for Motorola equipment.
Fig. 1. Network Topology.
2 1.5 1 0.5 0
1
2 3 Velocidad de Transmisión
4
Fig. 2. Bitrate performance at 1, 2, 3 and 4 Mbps. TABLE I. Rate [Mbps]
BITRATE RESULTS
1 Mbps
2 Mbps
3 Mbps
4 Mbps
TCP
1.024
2.047
3.071
3.968
TCP 15
1.023
2.045
3.071
3.968
TCP 20
1.023
2.047
3.071
3.507
IV.
RESULTS
Once the wireless link between Santa Cruz and San Cristóbal had been implemented, we proceeded to carry out the necessary tests in order to evaluate the performance of TCP over long distances wireless links. The parameters analyzed to assess the quality of service of the network were: Bitrate, jitter and delay. A. Bitrate Fig. 2 and Table I show the results of the TCP over wireless link. When TCP protocol was used under normal conditions, then the bitrate was similar to the traffic injected. Table I establishes the original TCP protocol is slightly better than the modified TCP protocol. It is also shown that at 4 Mbps, the modified TCP NACK+20 presents less reliable results than the other 2 evaluated protocols. Fig. 2 indicates that the performance of the generic TCP protocol, and the modified were similar until it reaches 3 Mbps. At 4 Mbps, TCP NACK+20 decreases the performance and the bitrate starts to fall. According to the data obtained from the equipment installed, its sensibility is between -98 dBm and -58 dBm, while the reception power is at -71 dBm. Under those conditions, the
expected throughput was 9.69 Mbps; therefore, the network efficiency was calculated to be around the 41.28% as shown in the next equation: !.!"[!"#$]
200
×100% = 41.28 %. 150
B. Jitter The results obtained are favorable in terms of the jitter, as depicted in Fig. 3. The TCP NACK+15 performs better than the generic TCP. Table II shows the jitter at the three transmission rates, with each TCP protocol. It is observed that the performance of the jitter decreases inversely with the transmission rate until it reaches 4 Mbps; at this point, the jitter value rises and generates a less stable network. Theoretically jitter should decrease as the bitrate increases. This behavior is observed in Table II until the network reaches 3 Mbps. After this, the value of jitter becomes larger at 4 Mbps for all the TCP versions. This comportment could be explained by the fact that the network reaches its physical limitations. Therefore, it can be said that TCP NACK+15 outstands the performance in terms of the jitter against the classic TCP protocol. Table III demonstrates that TCP NACK+15 has a better performance for all of the bitrates utilized.
Density
![!"#$]
𝜂=
TCP 15 Gaussiana
100
50
0
1.3036
1.3036
1.3036
1.3036 Data
1.3036
1.3036
1.3036 4
Fig. 5. Histogram delay at 3 Mbps in TCP 15
TABLE III.
x 10
IMPROVEMENT IN TERMS OF JITTER
Rate [Mbps]
1
2
3
4
TCP 15
4.74%
4.65%
3.22%
1.63%
TCP 20
4.74%
8.14%
0.85%
-14.05%
TABLE IV.
DELAY RESULTS
-3
5
x 10
TCP TCP 15 TCP 20
4.5 4 3.5
Jitter
3 2.5
Rate
1 Mbps
2 Mbps
3 Mbps
4 Mbps
TCP (s)
0.260947
0.334969
0.082168
0.304593
TCP 15 (s)
0.007398
0.0748808
0.003413
0.310491
TCP 20 (s)
0.011481
0.008992
0.003075
0.571323
2 1.5
C. Delay Delay is one of the main parameters for evaluating the network performance, and synchronization between PCs is essential. The end-to-end delay for each TCP protocol and for each bitrate as depicted in Fig. 4. Meanwhile, Fig. 5 demonstrates that delay follows a Gaussian patron. Delay between the PC’s was calculated by using histogram to measure the standard deviation. The greater the standard deviation is, the greater the dispersion of delay times. As shown in Table IV, results demonstrate that TCP NACK+15 is better suited for the network conditions. This protocol achieves a performance improvement in terms of the delay until 3 Mbps, but when the network starts to be saturated, the performance is much closer to the classic TCP.
1 0.5 0
1
2 3 Velocidad de Transmisión
4
Fig. 3. Jitter performance at 1, 2, 3 and 4 Mbps. 0.7 TCP TCP 15 TCP 20
0.6
0.5
Delay
0.4
0.3
0.2
0.1
0
1
2 3 Velocidad de Transmisión
4
Fig. 4. Delay Performance at 1, 2, 3 and 4 Mbps. TABLE II.
JITTER RESULTS
Rate Mbps]
1 Mbps
2 Mbps
3 Mbps
4 Mbps
TCP (s)
0.00464
0.00258
0.001644
0.002028
TCP 15 (s)
0.00442
0.00246
0.001591
0.001995
TCP 20 (s)
0.00442
0.00237
0.00163
0.002313
Even though the TCP NACK+20 had a better performance at lower bitrates, when the network was saturated this protocol had an abrupt increment of the delay. The percentage of gain in terms of delay for each bitrate is shown in Table V. According to the results obtained, they reveal the need to implement a cross layer communication in order to achieve a better performance of the TCP protocol. The modified TCP attempts to distinguish the type of lost packets, and to have a better reaction depending on that type. D. The Proposal TCP detects a lost packet either by the expiration of the RTO timer (TO), or by receiving three ACKs duplicated (TD).
In this sense, TCP reacts in a certain way for each type of lost packet according to the congestion avoidance procedure. The main issue for a long distance link is that RTO may expire before the sender could receive the corresponding ACK packet concluding in a poor performance of the Throughput, as shown in Fig. 6. In order to take advantage of the bandwidth available of the wireless link, the RTT of the transport layer relies on values determined in the link layer as described in [14], affecting the throughput performance. With the purpose of having a better cross layer communication, and an improvement of the TCP performance in terms of Throughput, it is needed an estimation of an appropriate RTO value according to the system conditions. Based on the mathematical model made in [14], RTT can be calculated in order to estimate RTO. In this model, the main equations were equations 1 and 2 of the TCP throughput model of TCP Reno described in [1]. !
EB ≈ ! !""
!!" !!" ! !!! × !"# !.! ! ×! × !!!"! !
B 𝑅𝑇𝑇, 𝑝 = min
!!"# ![!""]
, E[B] .
,
(1) (2)
Where E[B] denotes the mean number of TCP throughput. E[RTT] denotes the mean number of RTT; b denotes the number of segments with each acknowledgement acknowledged. T0 denotes the initial timeout. p denotes the segment loss probability. Wmax denotes the maximum congestion window size In conclusion, based on the modeling information, [14] found a relation between the throughput and contention window size on the MAC layer. Changging the contention window size affects the RTT and eventually impacts TCP throughput. Table VIII shows the RTT calculated and the RTO estimated for each bitrate. In that table, N stands for the number of frames of each TCP packet after fragmentation.
TABLE V.
ESTIMATED RTT AND CALCULATED RTO
RATE [Mbps]
Calculated RTT
1 Mbps
0.0123𝑁 + 5.587×10!! 𝑠
2 0.0123𝑁 + 5.587×10!!
𝑠
2 Mbps
0.0062𝑁 + 2.867×10!! 𝑠
2 0.0062𝑁 + 2.867×10!!
𝑠
3 Mbps
0.0041𝑁 + 1.961×10!! 𝑠
2 0.0041𝑁 + 1.961×10!!
𝑠
4 Mbps
0.0031𝑁 + 1.507×10!! 𝑠
2 0.0031𝑁 + 1.507×10!!
𝑠
V.
Calculated RTO = 2(RTT)
DISCUSSION
TCP is unable to distinguish the real cause of packet loss, it can be due to channel errors, transmission errors, collisions, or large propagation delays; causing a reduction of the bitrate. In this paper, we evaluated the performance of along distance wireless network, by using the proposal of TCP NACK developed in [12] and [13]. The procedure included a previous study of the state of the art of TCP, technical features of the equipment, network planning, and implementation of the network. The algorithm evaluated allowed an extension of the congestion window size of 15 and 20 segments, each time it received a negative ACK before congestion avoidance started. This paper describes a new algorithm, which improves the performance end to end for a long distance wireless link; it avoided unnecessary reductions of the congestion window size based on mathematical modeling. Results showed that TCP NACK+15 performed better until 3 Mbps, but when the bitrate reached its maximum at 4 Mbps, the throughput performance was decreased. It was noted the need to optimize the sensing mechanism of the error type to improve wireless TCP performance using cross layer design. Future work will include an optimization for communication between layers, to maximize the efficiency of the protocol, based on mathematical modeling described in this report, which could be achieved in an analytical and practical way We are interested in testing the algorithm under transmissions with specific applications that allows for practical purposes and use of their advantages to wireless telecommunications, overall in applications, which need realtime requirements as: VoIP or gamming over the Internet. ACKNOWLEDGMENT The authors gratefully acknowledge the contribution of Universidad de las Fuerzas Armadas (ESPE) for the financial support in the development of this project, through the Economical Support in the Development of this Project by Research under Grant Project 2011-PIT-010. This work might not have been possible had it not been for the efforts of Research Center of Ad-Hoc networks (CIRAD).
Fig. 6. Schematic timing diagram.
REFERENCES [1]
J. Padhye, V. Firoiu, D. F. Towsley y J. F. Kurose, «Modeling TCP reno performance: A simple model and its empirical validation,» IEEE/ACM Transactions on Networking, vol. 8, pp. 133-145, 2000. [2] E. Ayanoglu, S. Paul, T. F. LaPorta, K. K. Sabnani and R. D. Gitlin, "AIRMAIL: A Link-Layer Protocol for Wireless Networks," Journal Wireless Networks, vol. 1, no. 1, pp. 47-60, 19995. [3] H. Balakrishnan, S. Seshan y R. H. Katz, «Improving Reliable Transport and Handoff Performance in Cellular Wireless Networks,» ACM Wireless Networks , vol. 1, nº 4, pp. 469-481, 1995. [4] C. Parsa y J. Garcia-Luna-Aceves, «Improving TCP performance over wireless networks at the link layer,» Mobile Networks and Applications, vol. 5, nº 1, pp. 57- 71, 2004. [5] M. Mathis, J. Mahdavi, PSC, S. Floyd, LBNL y A. Romanow, «TCP Selective Acknowledgment Options,» The Internet Engineering Task Force (IETF), 1996. [En línea]. Available: http://www.ietf.org/rfc/rfc2018. [Último acceso: 10 10 2013]. [6] M. Mathis y J. Mahdavi, «Forward acknowledgement: refining TCP congestion control,» ACM SIGCOMM Computer Communication Review, vol. 26, nº 4, pp. 281-291, 1996. [7] S. Keshav y S. Morgan, «SMART retransmission: performance with overload and random losses,» INFOCOM '97. Sixteenth Annual Joint Conference of the IEEE Computer and Communications Societies. Driving the Information Revolution., Proceedings IEEE, vol. 3, pp. 1131 - 1138, 1997. [8] L. S. Brakmo, S. W. O’Malley y L. L. Peterson, «TCP Vegas: new techniques for congestion detection and avoidance,» ACM SIGCOMM Computer Communication Review, vol. 24, nº 4, pp. 24-35, 1994. [9] C. P. Fu y S. Liew, «TCP Veno: TCP Enhancement for Transmission over Wireless Access,» IEEE Journal on Selected Areas in Communications, vol. 21, nº 2, pp. 216- 228, 2003. [10] S. Mascolo, C. Casetti, M. Gerla, S. S. Lee y M. Sanad, «TCP Westwood: congestion control with faster recovery,» UCLA CSD Technical Report #990017, Los Angeles, 2000.
[11] K. Xu , Y. Tian y N. Ansari , «TCP-Jersey for wireless IP communications,» IEEE Journal on Selected Areas in Communications, vol. 22, nº 4, pp. 747-756, 2004. [12] G. Olmedo, «Controle de Congestionamento do Protocolo TCP em Sistemas de Comunicação sem fio CDMA usando estratégias de detecção multiusuário, arranjo de antenas e correção de erro FEC,» UNICAMP Universidad Estatal de Campinas, Brasil, 2008. [13] P. Pilo-País, «Verificación del desempeño de un nuevo algoritmo de control de congestionamiento en entornos inalámbricos reales mediante la modificación del protocolo TCP en el kernel de LINUX,» ESPE Escuela Politécnica Del Ejército, Sangolquí, 2011. [14] H. Xie, R. W. Pazzi y A. Boukerche, «A Novel Cross Layer TCP Optimization Protocol over Wireless Networks by Markov Decision Process,» IEEE Global Communications Conference (GLOBECOM), 2012. [15] J. Ye, J. Wang, Q. Liu y Y. Luo, «An Improved TCP with Cross-layer Congestion Notification over Wired/wireless Hybrid Networks,» The 9th International Conference for Young Computer Scientists, 2008. ICYCS 2008. , pp. 368-373, 2008. [16] D. K. Sung, N. N. Li, W. B. Zhu y X. M. Zhang, «TCP Congestion Window Adaptation Through Contention Detection in Ad Hoc Networks,» IEEE Transactions on Vehicular Technology, vol. 29, nº 9, pp. 4578-4588, 2010. [17] Lara. R, Olmedo. G, Calvopiña. K, «Performance evaluation of a New Wireless TCP Algorithm in out-door and in-door environments on WiFi and WiMAX links,» IEEE Chilecon, 2015. [18] A. Botta, A. Dainotti y A. Pescap, «Multi-protocol and Multi-platform Traffic Generation and Measurement,» Universita di Napoli Federico II, Napoli, 2011. [19] Manual SIRENET, Manual Sirenet versión 3.0, Espana: Intelia Consultores S.L., 2007. G. Eason, B. Noble, and I.N. Sneddon, “On certain integrals of Lipschitz-Hankel type involving products of Bessel functions,” Phil. Trans. Roy. Soc. London, vol. A247, pp. 529-551, April 1955.
