Aggressive Congestion Control Mechanism for Space Systems
Jingjing Wang, Chunxiao Jiang, Haijun Zhang, Yong Ren, Victor C. M. Leung
aa r X i v : . [ c s . N I] J a n Aggressive Congestion Control Mechanism forSpace Systems
Jingjing Wang, Chunxiao Jiang,
Member, IEEE , Haijun Zhang,
Member, IEEE ,Yong Ren,
Member, IEEE , and Victor C. M. Leung,
Fellow, IEEE
Abstract
How to implement an impeccable space system-of-systems (SoS) internetworking architecture hasbeen a significant issue in system engineering for years. Reliable data transmission is considered one of themost important technologies of space SoS internetworking systems. Due to the high bit error rate (BER),long time delay and asymmetrical channel in the space communication environment, the congestioncontrol mechanism of classic transport control protocols (TCP) shows unsatisfying performances. With thehelp of existing TCP modifications, this paper contributes an aggressive congestion control mechanism.The proposed mechanism is characterized with a fast start procedure, as well as the feedback informationto analyze network traffic and with a link terminating processing mechanism, which can help to revealthe real reason of packet loss, and maintain the size of congestion window at a high level. Simulationresults are shown in the end to verify the proposed scheme.
Index Terms
System engineering, space systems, congestion control mechanism, TCP modification, reliable datatransmission.
I. I
NTRODUCTION
Space system of systems (SoS) [1] is becoming a key consideration in the space communication,navigation, earth observation, etc. The space SoS is defined as a network of assets on the Earth, inthe orbit around the Earth, in the orbit around the solar system bodies, and on the surface of solar
J. Wang, C. Jiang, and Y. Ren are with the Department of Electronic Engineering, Tsinghua University, Beijing, 100084,China. E-mail: [email protected], [email protected], [email protected]. Zhang and V. C. M. Leung are with the Department of Electrical and Computer Engineering, The University of BritishColumbia, Vancouver, BC V6T 1Z4, Canada (e-mail: [email protected]; [email protected]). system bodies that is inter-connected and/or inter-operated to perform a mission, and/or provide servicesthat cannot be performed by monolithic space systems alone [2]. However, current space informationsystems are basically operated in a point-to-point pattern between the control center and the spacecrafts,which are only adequate for the individual space missions rather than collaborative missions. In otherwords, the ground network, space network and the deep space network evolve independently and focuson their individual communication regimes. To address these issues, the National Aeronautics and SpaceAdministration (NASA) has developed an integrated space system that integrates the Internet protocols,routers and interfaces into space networks [3] [4]. The space system, which is also called space Internet,enables large quantities of collaborative operations of those networks. Therefore, the integrated spaceInternet is essentially an SoS.With more frequent and complex aeronautics and astronautics missions recently, as one of the mostimportant technologies of space SoS internetworking systems, reliable data transmission has drawn moreand more attention. It is because that costs and risks grow with the increasing of the number of linksand cross-links. Therefore, in order to reduce the system risks and lower the design costs, it is necessaryto establish a reliable and efficient data transmission protocol frame in space Internet. Considering thechallenging communication environment in the space networks, the major differences between the spacenetwork and the ground network are summarized as follows [5]. a) High BER: Due to the multilevelatmosphere, uncertain weather conditions and the high-speed movement of communication terminals,typical BER in the satellite communication is from − to − . b) Long time delay: Because ofthe long communication distance, the round trip time (RTT) is beyond imagination. In most cases, thepropagation delay from the ground station to the low earth orbit (LEO) spacecraft is about 25ms, 110msto the medium earth orbit (MEO) and 250ms to geostationary earth orbit (GEO). Even worse, in terms ofthe interplanetary network, the one-way propagation delay is up to seconds even minutes. c) Asymmetricchannel: In the space system, the ration of downlink bandwidth to uplink bandwidth is over .d) Intermittent connection: The high-velocity motion of spacecrafts makes the communication terminalsbe out of sight periodically. During this period, connections are interrupted and communications aresuspended.The aforementioned challenges in space systems make it difficult to keep using the ground Internettransport control protocols to realize reliable data transmission. In [6], the classical TCP was a connection-oriented, duplex and reliability-ensuring protocol. It guaranteed the reliable data transmission mainly bythe congestion control mechanism. TCP-Tahoe [7] and TCP-Reno [8] were the two successful versionsbased on scaling-windows applied in the Internet protocol structure. However, the performances of both versions were deficient and unsatisfying in the space communication environment, where it is the highBER instead of network congestion that leads to the packet loss. Many other modified transport controlprotocols were also proposed for space systems. In [9], the authors summarized the existing modifiedprotocols into three categories. The first category allowed the modification only at the end terminals,like TCPW [10]. In the second category, both transport control mechanism and intermediate nodes wereimproved. In this case, the TCP connection was usually divided into some sub-connections with themethods of TCP-splitting and TCP-spoofing. In the third category, the protocols could achieve the reliabledata transmission function in the convergence layer of delay-tolerant network (DTN) [11] model. Inthe following, this paper surveys and introduces the existing modified transport control protocols andcorrespondingly lists their limitations [12].In Table I, we summarize the advantages and disadvantages of classical transport protocols [13] [14][15]. Among these multifarious modified protocols, some were in the testing phase emulating the realscenarios, while others were merely in the numerical analysis phases. Moreover, it is worth mentioning thatDTN-based space technologies have attracted more researchers’ attention in recent years. In [16][17][18],the authors proposed some data transmission protocols based on DTN, especially for deep-space com-munications. Moreover, the DTN-based data transmission protocols obtained a superior performancesagainst TCP-based data transmission protocols under the long-delay and asymmetrical channel. BecauseTCP-based transmission protocols rely on the timely feedback of acknowledgments(ACKs), while thelong-delay feedback and constrained uplink capacity result in a large number of ACKs to be lost.However, the operation of DTN-based data transmission protocols depend heavily on the bundle protocolto form a store-and-forward overlay network, which needs a cross-layer design and complex configurationin the terminals even the routers. Furthermore, the connectionless property of DTN-based protocolsbrings great challenges in the security issues of information transmission. Consequently, there is not acomprehensive or efficient version towards space network transport control protocols, let alone a space-ground integrated protocol stack. Therefore, we need urgently a simple, compatible and reliable congestioncontrol mechanism. Furthermore, from the related work, we can summarize that the modifications shouldobserve the following rules: a) The redundancy of a data segment should be compressed as much aspossible. b) The congestion control mechanism and error control mechanism can adapt to the long-delay,high BER, asymmetry and intermittent interruption channel environment. c) Take full advantage of thegateways or routers to forward data, if necessary. To the greatest extent, we should keep the originalrouting protocol unchangeable or less changeable.Therefore, to address the aforementioned issues, we propose an aggressive congestion control mech- TABLE IA
SUMMARY ON TRANSPORT CONTROL PROTOCOLS FOR SPACE NETWORK
Protocols Characteristics LimitationsTCP-Peach Better for long delay Ignore high BERTCP-W Check network traffic Useless in long delayTP-Planet Establish two states Complex, incompatibleXCP IP record network condition Split TCP connectionPETRA Intermediate node to forward Lack of securityREFWA Better efficiency and fairness Ignore high BERSCPS-TP Multiple improvement Complex, incompatibleBP Cope with DTN Cross-layer designLTP Hierarchical data transmission Cross-layer designanism, aiming at obtaining a higher link utilization rate and a larger throughput in data transmitting.This mechanism should be configured based on the traditional TCP protocol architecture and the datapackets are transparent over intermediate nodes. Moreover, it is appropriate for the space communicationenvironment network with comprehensive packet loss rate under 5% as well as low end-to-end latency,while a large end-to-end delay and frequent connection interruptions are tolerant in the DTN protocols.To the best of our knowledge, the congestion control mechanism can well accommodate to the spacecommunication environment under the hypothesis mentioned above. Besides, the proposed mechanismhas the characteristics of simplicity, low cost, strong operability and easy implementation. Moreover, itobviously improves the link utilization and has a commendable performance in robustness and compati-bility. The reason of calling it ”aggressive” is that the congestion control mechanism recovers the size ofsending window rapidly, maintains the congestion control window at a high level and has a remarkableperformance in the high BER environment.The rest of this paper is organized as follows. We first propose and describe an innovative congestioncontrol mechanism for space systems in Section II. In Section III, we demonstrate the simulation resultsand some performance analysis. Finally, conclusions are drawn in Section IV.II. C
ONGESTION C ONTROL M ECHANISM FOR S PACE S YSTEMS
In view of the existing modification methods which either lacked the compatibility with the groundInternet or excessively reduced the sending window, we propose a new ”aggressive” modification mech-anism with the following motivations. First, we should take full advantage of the characteristic of largebandwidth-delay produce (BDP) in the space communication environment. A feasible method tries toincrease the sending window and to change the classical TCP slow start phase. Second, cutting the congestion threshold down cursorily squanders the channel resources. We should make use of the historicalinformation to estimate the network traffic condition and smoothly reduce the congestion threshold. Third,the proposed modification can well manage the abrupt link interruption due to the high speed movementof aircrafts.
Fast start : Given the receivers feeding back one acknowledgement (ACK) for each successfully receiveddata segment, according to the principles of TCP slow start procedure, it needs t ss to reach the stabletransmission rate B . t ss can be calculated by t ss ≈ RT T · [1 + log ( B · RT T /l )] , (1)where RTT is the round trip time of each segment, which is about 50ms, 250ms and 550msin the scenarioof LEO, MEO and GEO, respectively. l equals 1KB in this paper, which represents the average bit lengthof data segments. If stable transmission rate B is 10Mbps, according to equation (1), in the GEO scenario, t ss is up to 5.73s; if the stable transmission rate B can reach 100Mbps, under this assumption, the value of t ss is 7.91s in the same scenario. Obviously, it takes too long to meet the need of general communicationmissions. A simple substitution of slow start named fast start is proposed in this paper. In the fast startstage, when senders deliver a data segment, they transmit an empty segment simultaneously. The emptysegment containing a compressed TCP header motivates the receivers to feed back ACKs, which largelyaccelerates the size of congestion window ( cwnd ). The size of senders’ cwnd grows as the speed of n rather than n . This speed-up mechanism is transparent to routers and receivers. In this respect, there isno need to modify other nodes in the network, which simplifies the modification operation and maintainsa good compatibility. Therefore, fast start is compatible with the real ground Internet and space networkprotocol stacks. Adaptive congestion control mechanism : When receiving 3 duplicate ACKs, instead of halving the cwnd ,senders utilize
RTT carried by ACKs to analyze the real reason of packet loss [19]. For different reasonssuch as high BER or network congestion, they take different actions. For making accurate judgements, weshould collect and analyze some parameters in the network. Let σ represent the variable of throughput,which can be calculated as follows: σ = ( Expected − Actual ) · base RT T, (2)where base RT T is the reference standard of RTT and variable
Expected = cwnd/base RT T . It isnotable that parameter Actual should be calculated accurately, because it directly reflects the present state of the network. However, the network states possess continuity and they can not change tempestuously.In consequence, the space system can be considered having memory characteristic. In order to obtain theaccurate value of
Actual , senders need to record the historical observation value
RT T i and recordingtime T i , where i shows the different measure moments. Then Actual can be determined by the followingexpressions:
Actual = cwnd/ ] RT T ] RT T = n X i =1 A i e − τ RT T i , (3)where attenuation index τ = T n − T i , i = 1 , , ...n , T n represents the present moment, and T is themoment when the terminals establish the connection. A i is the normalization coefficient.Furthermore, we set a threshold of σ with variable β . As mentioned above, σ approximately reflects thevariation per base RT T of the cwnd against expected value. During a full base RT T , the losing of threedata packets is indeed a small probability event under the condition of normal connection without networkcongestion. In this article, let congestion threshold β be a constant valued 3 in this paper. When σ ≤ β ,we regard the packet loss as a result of random bit error. Thus, the sender retransmits the lost packetimmediately. Meanwhile, the congestion threshold ( ssthresh ) maintains unchangeable and congestionwindow increases by three data segments due to the three duplicate ACKs received successfully, i.e., ssthresh = cwndcwnd = ssthresh + 3 . (4)On the other hand, if σ > β , we regard the packet loss as a result of network congestion. At this time, thesender retransmits the lost packet, while the ssthresh and cwnd are calculated as the following equations: ssthresh = cwnd × kcwnd = ssthresh + 3 k = βσ base RT T ] RT T , (5)where the threshold control coefficient depends on the degree of network traffic congestion. In this way, ssthress can be reset to a reasonable value. When the retransmission timer times out, similar to theTCP-Reno mechanism, senders halve the ssthresh and set cwnd to 1.
Congestion windows maintaining : The high speed movement of satellites leads to a high probabilityof link interruption. An abrupt increasing of
RTT causes a leap of σ . To make matters worse, the cwnd TABLE IIA
VERAGE HOLD - ON TIME AND NEW CALL BLOCKING RATE IN THE SECOND SCENARIO
Simulation Protocols Average hold-on time New call blocking rateNew TCP 1.948ms 0.012%TCP-Tahoe 2.216ms 0.028%TCP-Reno 2.031ms 0.018%TCP-Vegas 1.998ms 0.017%TCPW 1.962ms 0.014%will shrink persistently until the date transmission link rebinds. Under this circumstance, we propose acongestion window maintaining mechanism. First, we should estimate the distance of the whole TCPconnection. In [20], the instantaneous geocentric angle θ of two satellites is written as follows: θ = arccos[sin ϕ a sin ϕ b + cos ϕ a cos ϕ b cos( ψ a − ψ b )] , (6)where ( ϕ a , ψ a ) and ( ϕ b , ψ b ) are the sub-satellite point longitude-latitude coordinates of satellite a and b .Then the distance of communication link d can be calculated via: d = p ( r + h a ) + ( r + h b ) − r + h a )( r + h b ) cos θ, (7)where r is the earth radius, and h a and h b are the satellite orbital altitudes. Therefore, the total linkdistance D is the sum of each point to point distance d ij . Then the estimated RT T est = 2
D/c . The high-speed movement of two satellites leads to a corresponding change in the instantaneous geocentric angle θ . According to the distribution and the altitude of stationary orbit satellites, let us assume ≤ θ ≤ π .Based on this hypothesis, we conservatively take RT T est as the interruption threshold. When the
RT T i > RT T est , we can judge that the data transition link breaks off because of the high speedmovement. Simultaneously, maintaining the size of cwnd , holding the time-out clock, and recording thetransmitting state, the sender continuously dispatches detective packets. Once the communication linkreconnects, the data transmission recovers.This modified TCP congestion control mechanism can accurately distinguish the reason of packet loss,and can hold the congestion window to a large extent. In the space communication environment of alarge bandwidth delay product, this aggressive modification can expand the throughput and increase thelink utilization rate significantly.
GEO MEOLEO
1. 10M FTP Downloading2. Stochastic Call Access3. Self-similar VBR Video
Fig. 1. Simulation scenario and simulation business model
III. S
IMULATION R ESULTS A ND A NALYSIS
This paper uses the network simulation software OPNET to verify the performance of the newcongestion control mechanism. Fig. 1 demonstrates the space network simulation scenarios. The spacesystem is constructed by kinds of satellites in three layers denoted as GEO, MEO and LEO. The satellitesin the same layer can communicate with each other by means of a relay satellite in the higher layer.The red and blue arrow lines represent the data transmission between two satellites via a relay satellite.This process of data transmission is noted in the right corner of the figure. The data through a wire link,routers, wireless link and relay satellite is received by clients from servers. All the simulations are underthe circumstances with milliseconds end-to-end delay and comprehensive packet loss rate under 5%.In order to verify the performance of the modified protocol sufficiently, we simulate in different situ-ations with three business models. First, we simulate three 10MB FTP persistent downloading scenarioswith different packet loss rates, . , and , respectively. The three kinds of packet loss rates reflectdifferent round trip distance and various spatial environments in the real world. However, for obtainingan obvious result, the three packet loss rates are much higher than those in the ground Internet network.The simulation time is set to 20 minutes. During this period, clients can accomplish the FTP downloadingmission from servers normally. We compare the new mechanism with classical TCP congestion control
200 300 400 500 600 700 800 900 1000 11000123456789 x 10 Simulation time(sec) T h r oughpu t ( b it s / s ec ) New−0.5%Reno−0.5%Tahoe−0.5%Vegas−0.5%TCPW−0.5% (a) Time-average throughput in . packet lossrate
200 300 400 500 600 700 800 900 1000 1100012345678 x 10 Simulation time(sec) T h r oughpu t ( b it s / s ec ) New−1%Reno−1%Tahoe−1%Vegas−1%TCPW−1% (b) Time-average throughput in packet loss rate
200 300 400 500 600 700 800 900 1000 1100024681012 x 10 Simulation time(sec) c w nd New−0.5%Reno−0.5%Tahoe−0.5% (c) Real time cwnd in . packet loss rate
200 300 400 500 600 700 800 900 1000 1100024681012 x 10 Simulation time(sec) c w nd New−1%Reno−1%Tahoe−1% (d) Real time cwnd in packet loss rate
200 300 400 500 600 700 800 900 1000 110000.10.20.30.40.50.60.70.80.9
Simulation time(sec) U tili za ti on New−0.5%Reno−0.5%Tahoe−0.5%Vegas−0.5%TCPW−0.5% (e) Time-average channel utilization in . packetloss rate
200 300 400 500 600 700 800 900 1000 110000.10.20.30.40.50.60.70.80.9
Simulation time(sec) U tili za ti on New−1%Reno−1%Tahoe−1%Vegas−1%TCPW−1% (f) Time-average channel utilization in packetloss rate
200 300 400 500 600 700 800 900 1000 110000.020.040.060.080.10.120.140.160.180.2
Simulation time(sec) U tili za ti on New−5%Reno−5%Tahoe−5%Vegas−5%TCPW−5% (g) Time-average channel utilization in packetloss rate
200 300 400 500 600 700 800 900 1000 1100012345678 x 10 Simulation time(sec) T h r oughpu t ( b it s / s ec ) New VBRReno VBRTahoe VBRVegasTCPW (h) Time-average throughput in VBR video scenar-iosFig. 2. Simulation results in different channel situations and different business models (FTP downloading and VBR videotransmitting) mechanisms in the expects of average throughput and link utilization. Second, the business model ispurely stochastic. Fifty thousand INMARSAT telephone calls are accessed within 2 hours as a Poissonprocess with ≤ t ≤ and average arriving rate λ = 14 . In each hold-on time, senders transmit 500Bytes data to the receivers, and then hang off. We record and compare the average hold-on time of eachcall and the new call blocking rate. In the third situation, we consider the variable bit rate (VBR) videobusinesses which are not independent from each other. In other words, the variable bit rate (VBR) videobusinesses have the characteristic of long-range dependence. This self-similar property can be describedby the Pareto distribution as in (8), where k = 1 and α = 1 . . f ( x ) = αk α /x α +1 , x > k. (8)During the simulation time, senders transmit large amounts of long-range dependence VBR video. Andthe average throughput is recorded.From the subgraphs (a) ∼ (d) of Fig. 2, we can see that the new congestion control mechanism hasa better performance than others in the environment of high BER. It has a good robustness in severecircumstances. The higher the packet loss rate is, the more advantages the new mechanism has. Fromthe average-processing throughput curves and real time cwnd curves in the two situations with . and packet loss rate, we can conclude that the receiver with the new mechanism first completes the10MB FTP downloading mission, and maintains a higher throughput over classical Tahoe, Reno, Vegasand TCPW from the beginning to the end. The simulation results evidently reflect the superiority of thenew congestion control mechanism.Subgraphs (e) ∼ (g) of Fig. 2 show the link utilization rate of each mechanism with three packetloss rates. They show that the performance of link utilization is enormously promoted with the newaggressive congestion control mechanism in high BER environment. What is more, the new congestioncontrol mechanism achieves the best relative performance in the scenario with packet loss rate againstothers.From the Table II, we conclude that the new congestion control mechanism can well handle thestochastic business. With the help of the new congestion control mechanism, the hold-on time is shortenedevidently. In other words, it has the least time of data transmitting and the minimum probability of new callblocking against the other TCP protocols. Moreover, the simulation results of the third scenario is shownin the subgraph (h) of Fig. 2. Obviously, the modification mechanism possesses the best performance inthe average throughput against others in the VBR business model. IV. C
ONCLUSION
This paper presented a novel congestion control mechanism for space systems. The proposed mech-anism which makes the size of cwnd maintain a large value fully adapts to the space informationenvironment of intermittent interruption and high BER. Because of the new congestion control mechanismbeing able to detect the real reason of packet loss, it can well regulate the network traffic and recoverthe throughput rapidly. Furthermore, the simulation results showed that the aggressive mechanism had abetter performance over Reno, Tahoe, Vegas and TCPW in space communication environment. Moreover,the higher the BER is, the more obvious advantages of the new mechanism over traditional TCPs are.Compared with other well-performed modification versions, the novel modified mechanism is easier to beimplemented and configured. We only need to amend the transport control protocol on the transmissionside, and it is transparent to all intermediate routers and receivers.R
EFERENCES [1] M. Jamshidi,
System of systems engineering: innovations for the twenty-first century . John Wiley & Sons, 2011, vol. 58.[2] K. B. Bhasin and J. L. Hayden, “Communication and navigation networks in space system of systems,”
System of SystemsEngineering , pp. 348–384, 2008.[3] J. Hu, R. Wang, Q. Zhang, Z. Wei, and J. Hou, “Aggregation of DTN bundles for space internetworking systems,”
SystemsJournal, IEEE , vol. 7, no. 4, pp. 658–668, 2013.[4] K. B. Bhasin and J. L. Hayden, “Architecting communication network of networks for space system of systems,” in
Systemof Systems Engineering, 2008. SoSE’08. IEEE International Conference on . IEEE, 2008, pp. 1–7.[5] A. R. Urke, “Transport layer challenges in hybrid military satellite networks,” 2011.[6] K. R. Fall and W. R. Stevens,
TCP/IP illustrated, volume 1: The protocols . addison-Wesley, 2011.[7] Z. Wang and J. Crowcroft, “Eliminating periodic packet losses in the 4.3-Tahoe BSD TCP congestion control algorithm,”
ACM SIGCOMM Computer Communication Review , vol. 22, no. 2, pp. 9–16, 1992.[8] J. Padhye, V. Firoiu, D. F. Towsley, and J. F. Kurose, “Modeling TCP Reno performance: a simple model and its empiricalvalidation,”
IEEE/ACM Transactions on Networking (ToN) , vol. 8, no. 2, pp. 133–145, 2000.[9] R. Wang, T. Taleb, A. Jamalipour, and B. Sun, “Protocols for reliable data transport in space Internet,”
CommunicationsSurveys & Tutorials, IEEE , vol. 11, no. 2, pp. 21–32, 2009.[10] C. Casetti, M. Gerla, S. Mascolo, M. Sanadidi, and R. Wang, “TCP Westwood: end-to-end congestion control forwired/wireless networks,”
Wireless Networks , vol. 8, no. 5, pp. 467–479, 2002.[11] X. Sun, Q. Yu, R. Wang, Q. Zhang, Z. Wei, J. Hu, and A. V. Vasilakos, “Performance of DTN protocols in spacecommunications,”
Wireless networks , vol. 19, no. 8, pp. 2029–2047, 2013.[12] R. Wang and S. Horan, “Protocol testing of SCPS-TP over NASA’s ACTS asymmetric links,”
Aerospace and ElectronicSystems, IEEE Transactions on , vol. 45, no. 2, pp. 790–798, 2009.[13] T. Taleb, N. Kato, and Y. Nemoto, “REFWA: an efficient and fair congestion control scheme for LEO satellite networks,”
IEEE/ACM Transactions on Networking (TON) , vol. 14, no. 5, pp. 1031–1044, 2006. [14] M. Marchese, M. Rossi, and G. Morabito, “PETRA: Performance enhancing transport architecture for satellite communi-cations,” Selected Areas in Communications, IEEE Journal on , vol. 22, no. 2, pp. 320–332, 2004.[15] M. Luglio, M. Y. Sanadidi, M. Gerla, and J. Stepanek, “On-board satellite split TCP proxy,”
Selected Areas inCommunications, IEEE Journal on , vol. 22, no. 2, pp. 362–370, 2004.[16] Q. Yu, R. Wang, Z. Wei, X. Sun, and J. Hou, “DTN Licklider transmission protocol over asymmetric space channels,”
Aerospace and Electronic Systems Magazine, IEEE , vol. 28, no. 5, pp. 14–22, 2013.[17] J. Hu, R. Wang, X. Sun, Q. Yu, Z. Yang, and Q. Zhang, “Memory dynamics for DTN protocol in deep-spacecommunications,”
Aerospace and Electronic Systems Magazine, IEEE , vol. 29, no. 2, pp. 22–30, 2014.[18] Q. Yu, R. Wang, and K. Zhao, “Performance modeling of LTP data transmission protocol in deep-space communications,”
Aerospace and Electronic Systems, IEEE Transactions on , vol. 51, no. 3, Jul. 2015.[19] C. P. Fu and S. C. Liew, “TCP Veno: TCP enhancement for transmission over wireless access networks,”
Selected Areasin Communications, IEEE Journal on , vol. 21, no. 2, pp. 216–228, 2003.[20] C. Pan, D. Wei, Y. Bie et al. , “Improvement on TCP vegas algorithm for satellite network,”