Improving Throughput of Starved TCP Flow by Sidestepping Bottleneck Nodes Using Concurrent...

68 International Journal of Mobile Computing and Multimedia Communications, 2(1), 68-83, January-March 2010

Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Globalis prohibited.

Keywords: ConcurrentTransmission,Congestion,Fairness,IEEE802.11MAC,MANETs,SANETs

InTroduCTIon

The traditional transport layer protocol TCP behaves very unpredictably with degraded throughput in multi-hop wireless ad hoc net-works, as it was designed to provide reliable

end-to-end delivery of data packets over wired networks. In wireless scenarios, the main prob-lem of TCP (Chen, Zhai, Wang & Fang, 2004; Fu, Meng & Lu, 2002) is the performance of congestion control mechanisms in case of losses which are not occurred due to network conges-tion. TCP is unable to distinguish between non-congestive losses (due to route failures because

Improving Throughput of Starved TCp Flow by

Sidestepping Bottleneck nodes using Concurrent Transmission

RajeshVerma,MNNIT,India

ArunPrakash,MNNIT,India

RajeevTripathi,MNNIT,India

NeerajTyagi,MNNIT,India

ABSTrACTTheTCPcongestioncontrolmechanismalongwithunfairnessproblemposespoorperformancewhenIEEE802.11MACprotocolisusedinmulti-hopadhocnetworksbecausethetraditionalTCPhaspoorinteractionwiththeIEEE802.11MACprotocol.BecauseofthegreedynatureofTCP,starvationproblemofTCPflowswithlongerpathsissevere.Inthisarticle,wefirstillustratethatthefairness,congestioncontrolandmediumcontention are closely coupled issues and the spatial reuse of the channel can improve the performance ofwirelessadhocnetwork.ByusingconcurrenttransmissionprotocolattheMAClayer,likeCTMAC,inmulti-hop networks we can achieve simultaneous transmissions within the interference regions. Further, we illustratewithextensivesimulationsinns-2thatbyschedulingmultipleconcurrenttransmissionsalongthepathlinks,thestarvationproblemduetogreedynatureofTCPcanbeeliminatedandensuinghigherflowthroughput and lower end-to-end delay.

DOI: 10.4018/jmcmc.2010103004

International Journal of Mobile Computing and Multimedia Communications, 2(1), 68-83, January-March 2010 69


of mobility, random errors, contentions on the wireless channel and unfairness) and due to network congestion. So, TCP performs poorly in wireless ad hoc networks because it has to face new challenges due to several causes specific to these networks such as lossy channels, hidden and exposed nodes, path asymmetry, network partitions & route failures (in multi-hop envi-ronments) and power limitations.

When IEEE 802.11 MAC protocol is used in the multi-hop ad hoc networks, the traditional TCP congestion control mechanism encounters several new problems. In these networks, the medium contention is high which results in severe unfairness and starvation problems for TCP flows. The congestion control, fairness and medium contention are all related and closely coupled issues for TCP end-to-end throughput and the end-to-end delay. Several works have been done in the past related to greedy nature of TCP resulting in high level of congestion and thus degrading the performance of the wireless ad hoc networks. For example, Figure 1 shows a simple 9 static node chain topology with a single TCP flow from node 1 to node 9 with 1000 bytes payload, it is observed that the TCP traffic introduces a lot of collisions and therefore retransmissions of control (RTS/CTS) and Data packets at the MAC layer [3] and there are many TCP packets dropped at a rate of 0.83 to 3.63 packets/sec due to medium contentions without any queue overflow.

Again, TCP traffic is unstable in the wireless multi-hop scenario because the round trip time (RTT) oscillates in a larger range and so does the instantaneous throughput. These observations are due to the greedy nature of TCP and the

strong dependencies between congestion and the medium contentions. The TCP increases the congestion window size till the first packet loss and when the transmission rate of the sender crosses the channel capacity, the subsequent incoming packets are accumulating at various nodes along the directed route. When all the neighboring nodes have packets in queue to transmit, they try for contending the channel and thus number of collisions increases and so is the channel contention delay, thereby slowing down the packet forwarding rate and frequent congestion occurs in the network. Thus congestion and the collision work jointly and continuous retransmission timeouts and delayed duplicate ACKs occurs giving poor end-to-end throughput and larger delays. Here in this simulation work, we will be using con-current transmission MAC protocol (CTMAC), which was derived for single-hop independent multiple concurrent transmissions in MANETs for increasing the spatial reuse of the channel within the interfering regions in our proposed static ad hoc networks (SANETs) chain topology scenario because of its simple implementation and better results.

The rest of the article is organized as fol-lows: In section II, we discuss the related work on solutions to starvation problem and concur-rent transmission protocols at the MAC layer level. We present a summary of IEEE 802.11 and working of a concurrent transmission protocol, CTMAC, which we are using in our proposed work, in section III. The motivation and prob-lem formulation is discussed in section IV. In section V, we describe the proposed solution to the congestion and starvation problem. Section

Figure 1. A 9-node topology with a single flow



VI discusses the simulation results followed by conclusion in section VII.

rElATEd work

Starvation problem of TCP flows can result in severe congestion and hence degraded through-put. Link-RED (Fu, Zerfos, Luo, Lu, Zhang & Gerla, 2003) was proposed to solve this problem where according to observed packet collisions, the TCP packets were marked or dropped. Subsequently the TCP source reduces or adjusts the congestion window limit dynami-cally according to the path length, to avoid the congestion. A neighborhood RED (XU. Gerla, Qi & Shu, 2003) scheme was proposed to deal with unfairness problem by adjusting the mark-ing or dropping probability depending on the channel information.

It is known that the spatial reuse of the chan-nel can be achieved by scheduling simultaneous transmissions whose regions are not in conflict. In the IEEE 802.11 MAC protocol, the minimum distance between two simultaneous transmitting nodes is of four hops as the interfering region is determined by the signaling signal range which is slightly greater than twice the transmission range of a node. There are several available concurrent transmission protocols for mobile ad hoc networks (MANETs) in literature, most of them are using single hop multiple transmissions for improving the throughput. Some of them are using Additional Control Gap (ACG) which is inserted between the RTS/CTS and DATA packets for scheduling the parallel transmis-sions. The neighboring nodes try successfully to access the channel within this ACG. In MACA-P protocol (Acharya, Misra & Bansal, 2003) and its enhancement (Acharya, Misra & Bansal, 2004), the tolerable Signal-to-Interference-and-Noise Ratio (SINR) during receiving periods of a node is not included so it restricts the parallel transmissions whenever the two neighboring nodes are a transmitter and a receiver, that is, neighboring sending and receiving sessions of different transmissions fail or there can not be a receiver (session 1) within the transmission

range of a sender (session 2). Also in MACA-P the ACG is fixed so there cannot be more numbers of parallel transmissions possible. Another protocol, CTMAC (Yu, Cao, Zhou, Wang, Keith C. C. Chan, Alvin T. S. Chan & Leong, 2008) has flexibility of increasing the ACG slots depending on the number of data ready senders thereby enabling multiple parallel transmissions. Here, due to the consideration of SINR and collision avoidance information in the control packets, the neighboring nodes of different transmissions can be a sender and a receiver. The limitation of CTMAC is that the slave transmission cannot send data after the master transmission has finished because of the possibility of the interference from the ACK packets of the primary transmission. There are several Transmission Power Control (TPC) protocols like POWMAC (Muqattash & Krunz, 2005) available for scheduling the concurrent transmission, but due to their large numbers of limitations, practical implementation is very complex and requires additional hardware/control packets.

Another protocol called Overlapped Carrier-Sense Multiple Access (OCSMA) protocol (Boppana & Shea, 2009) is defined for multi-hop ad hoc networks where MAC and PHY interaction have been considered for scheduling the concurrent transmissions. At the PHY layer, Multiuser Detection (MUD) technique is used to recover information from colliding packets at the receiver. This is a signal processing technique used at the PHY to enable a node to receive packets in the presence of other transmissions in its communication range. This multi-packet reception capability of the nodes at PHY provides greater spatial reuse in the network. Using Interference Cancellation (IC) for known interference may also improve network performance. This type of known-IC technique is used in this protocol to increase simultaneous transmissions in wireless ad hoc networks.

The IEEE 802.11 protocol limits the spatial reuse of the wireless channel by deactivating all the neighboring nodes to the sender and re-ceiver. However, by using directional antennas,



it is possible to schedule multiple simultaneous transmissions in the neighborhood. Again, due to the higher gain, directional antennas have a larger transmission range as compared to omni-directional antennas. This makes possible for the two distant nodes to communicate over a single hop, thus routes using directional antennas may require fewer hops than what may be possible using omnidirectional antennas. Using these two important features of directional antennas, namely, spatial reuse and higher transmission range, some researchers in the past have ad-dressed the challenges of using the directional antennas to improve the wireless channel utili-zation (Balanis et al., 2000; Bharghavan 1998; Gupta & Kumar, 2000; Royer, Lee & Perkins, 2000; Sanchez, Giles & Zander, 2001; Ward & Crompton, 1992; Zander, 1990). Although in ad hoc networks, literature related to the directional antennas is limited. Nasipuri, Ye, You & Hiromoto (2000) addressed the problems related with the use of directional antennas in the design of a MAC protocol where they assumed that the gain of directional antennas is equal to that of omnidirectional antennas. Ko, Shankarkumar and Vaidya (2000) also assumed that gains of directional and omnidi-rectional antennas are equal, but they observed that when using directional antennas, a node may transmit in some specified directions that do not interfere with the existing transmis-sions. Bandyopadhyay, Hausike, Horisawa and Tawara (2001) presented a complex MAC protocol which uses extra messages to inform neighboring nodes about ongoing transmissions, thereby increasing the MAC layer control over-head. Takai, Martin, Bagrodia and Ren (2002) proposed directional virtual carrier sensing (DVCS) and directional network allocation vector (DNAV) mechanisms. Choudhury, Yang, Ramanathan and Vaidya (2002) discussed two MAC protocols used for directional antennas which exploit both the benefits of directionality. The nodes are working in two modes, namely, omnidirectional mode and directional mode. The basic directional MAC (DMAC) protocol is discussed to improve the spatial reuse of the channel and the multi-hop RTS MAC (MMAC)

protocol that exploits the extended transmission range capability of directional antennas by at-tempting to form longest possible links. This is achieved by transmitting an RTS over multiple hops and then consequently transmitting CTS, DATA, and ACK packets directionally over a single hop.

prElIMInArIES

IEEE 802.11

The MAC layer of IEEE 802.11 (IEEE, 1999) uses DCF (Distributed Co-ordination Function) as the basic Media Access Control protocol for MANETs. It is based on CSMA/CA (Carrier Sense Multiple Access with Collision Avoid-ance), with extensions to RTS/CTS. In this protocol, control signals RTS/ CTS are being sent over the channel prior to any DATA packet transmission. When a node is ready to send any data packet, it first broadcasts an RTS packet to its neighboring nodes; the intended node (receiver node) responds this RTS packet by a CTS packet. On hearing positive CTS packet, the sender starts transmitting data packets after DCF inter-frame space (DIFS) interval. When all of the data packets have been sent, receiver node acknowledges the sender about the suc-cessful reception by sending an ACK packet. Within these RTS/CTS packets, a field called NAV (Network Allocation Vector) is added to indicate the expected time of transmission/ reception. By hearing these RTS/ CTS packets (NAV field), other nodes within the transmission range of sending and receiving nodes, restrict themselves from contending for the channel thereby making the transmission successful. The IEEE 802.11 uses back-off interval to resolve the channel contention. Before starting the transmission of DATA packets, the sender node chooses a random back-off period varying in the range (0, CW), where CW is the Conten-tion Window (Chen, Xue & Nahrstedt, 2003). The sender node then starts decrementing the back-off timer by 1 as soon as it encounters an idle time slot. When it reaches 0, sender starts



transmitting DATA packets. If this transmis-sion collides with some other transmission, the sender doubles its CW, chooses a new back-off period and attempts retransmission. The CW is doubled each time it senses collision until it reaches some maximum value, CWMax. Besides its simple implementation, DCF is restrictive in nature as it prohibits other nodes (willing to transmit) from accessing the channel in spite of the possibility of simultaneous transmissions.

In ad hoc networks nodes rely on physical carrier sensing mechanisms to determine an idle channel as in IEEE 802.11 DCF function and there are two well known problems at the MAC layer namely: hidden-terminal and exposed-terminal problems. A typical hidden node sce-nario is presented in Figure 2, where Terminals X and Z have a packet to transmit to terminal Y. Terminal X cannot detect Z’s transmission because it is outside the transmission range of Z. Terminal Z (resp. X) is therefore “hidden” to terminal X (resp. Z). Since the transmission areas of X and Z are not disjoint, there will be packet collision at Y. These collisions make the transmission from X and Z to Y problematic. To alleviate the hidden terminal problem, virtual carrier sensing has been introduced which is a two-way handshaking that occurs before any data transmission. Two-way handshaking is done using a control frame transmission called request-to-send (RTS) from the sender

and another control frame transmission called clear-to-send (CTS) from the receiver. Both RTS and CTS frames contain the total duration of the data transmission. However the hidden station problem may persist in IEEE 802.11 ad hoc networks even with the use of the RTS/CTS handshake, because the power needed to interfere the packet reception is much lower than that required to deliver a packet successfully or in other words a node’s transmission range is much smaller than the node’s sensing range.

The exposed terminal problem results from a situation where a transmission between two terminals causes another transmission to be de-layed. In Figure 3, we show a scenario where the exposed terminal problem occurs. Let us assume that W and Y are within the transmission range of terminal X, and W is outside Y’s transmission range. Let us also assume that X is transmitting to W, and Y has a packet to be transmitted to Z. Due to carrier sense mechanism, terminal Y senses the channel as busy because of X’s transmission. Therefore, terminal Y will be restricted from transmitting to Z, even though this transmission will not cause interference at W. Thus exposed terminal problem may result in under utilization of the channel resources. The hidden terminal and exposed terminal problems are correlated with the transmission range as by increasing the transmission range, the hidden terminal problem occurs less frequently but

Figure 2. A hidden terminal problem



the exposed terminal problem becomes more prominent as the transmission range identifies the area affected by a single transmission.

Basic operation of CTMAC protocol

The CTMAC protocol has been used for single hop transmission and has been derived for in-creasing the throughput of MANETs through scheduling of concurrent transmissions at the MAC layer. It works with single-channel, single-transceiver and single-transmission power to increase the network throughput while preserv-ing the collision avoidance property of the IEEE 802.11 scheme. The spatial reuse of the channel can improve the throughput of a wireless ad hoc network and can be achieved by allowing simultaneous transmissions. By using concur-rent transmission protocols, like CTMAC, we can achieve simultaneous transmissions within the interference regions and thus increasing the spatial reuse of the channel. To see the basic operation of CTMAC, let us take an example as shown in Figure 4 and timing information in Figure 5. Node 2 first transmits an RTS packet to node 1, having information of the start times of node 2’s DATA packet (Tij

data) and 1’s ACK packets (Tij

Ack), where i and j may be transmitter and receiver respectively or vice-versa. Node 1 responds with a CTS packet to node 2, including similar timing information. This information is necessary so that a node in the neighborhood of node 1 or 2 can evaluate whether it can receive a data packet from some other node simultane-ously while node 2 is transmitting data to node

1. After the exchange of these control packets, node 2 does not immediately sends data packet, as in IEEE 802.11, but rather wait for an interval specified by additional control gap (ACG). The ACG consists of an adjustable number of access slots (AS). The duration of an AS is fixed and is equal to the sum of the time required to transfer the RTS, CTS, and ATS (abort to send or adjust to send) packets, plus the maximum back-off time (CWmax). The minimum value for ACG is one Access Slot, which allows one potential secondary transmission to be scheduled. In CTMAC the maximum value for ACG is set to 3 access slot, which allows at most three second-ary transmissions to be scheduled concurrently with the primary transmission.

The ATS packet is used by a secondary transmission to inform the neighboring nodes for necessary information. ATS has two mean-ings according to the two situations when it is used. The first situation is when the secondary receiver wants to change the values of Tdata or Tack scheduled by the secondary transmitter. It includes the new values in the CTS packet, so the secondary transmitter uses ATS (adjust to send) packet to inform its neighboring nodes of the adjusted values. In second situation, when it is not possible for the secondary receiver to get the data packet from the secondary transmitter. The secondary receiver replies with a nega-tive CTS packet, so the secondary transmitter uses ATS (abort to send) packet to inform the neighbors of the cancellation of the proposed transmission. During the ACG period, node 3 and node 4 can exchange control packets and

Figure3.Anexposedterminalproblem



schedule their transmission if possible. If the transmission between node 3 and 4 is success-fully scheduled, then the two transmissions can be done concurrently and the starting time of node 2’s and node 3’s data packet will be same. There is one strict requirement in CTMAC that the DATA packet of a secondary transmission (here 3→4) must be less than or equal to the DATA packet of corresponding primary trans-mission (2→1), such that there is no collision between ACK packet of primary transmission and DATA packets of secondary transmission. This is done by sequencing the ACK packets of concurrent transmissions as shown in Figure 5. Thus, only the interference between DATA packets of primary and secondary transmissions needs to be considered. Primary transmission is the one who first accesses and occupies the

channel followed by one or more secondary transmissions. In wireless ad hoc networks (WANETs), a packet can be received success-fully even in the presence of other overlapping or interfering packets, only if its instantaneous power is larger than the instantaneous joint interference power by a minimum certain threshold factor (Kim & Lee, 1999).This effect is called capture effect, and the threshold fac-tor is Signal-to-Interference-and-Noise Ratio (SINR). One node can receive packet success-fully as long as its required SINR is fulfilled. Therefore in Figure 4, without considering the ACK packets, as long as the required SINR of both node 1 and 4 is kept, then the DATA packets of transmissions 2→1 and 3→4 can be transmitted simultaneously.

Figure4.OperationalscenarioforCTMAC

Figure5.TiminginformationforoperationofCTMAC



MoTIvATIon And proBlEM ForMulATIon

Medium contention is one of the main causes of unfairness problem. If there are several flows then they may traverse through different geo-graphical regions and having different levels of contention due to number of contending nodes in that region and thus require different allocation of shared bandwidth of the wireless channel. Again, the flow with longer path (in terms of number of hops) consumes more network re-sources than the flows with short path lengths and is encountered with more contentions and collisions, resulting in more packet drops. In multi-hop ad hoc networks, the unfairness problem suffered by TCP flows may result to the starvation of one of the TCP flows. The hid-den and exposed terminal problems along with the greediness of TCP flows contribute to the starvation of flows and increased collisions in the network. The greedy nature of TCP can be explained with the help of Figure 6, assuming that there are two separate TCP flows passing through the links 1 to 2 and 4 to 5. We know that in chain topology, the maximum channel utilization is achieved by scheduling the nodes four hops away to transmit simultaneously as the carrier sensing range is slightly larger than twice of the transmission range of a node, so when node 4 is transmitting packets to node 5, node 2 senses the channel as busy and can-not respond to node 1. Now node 1 will be retransmitting the RTS packets several times to node 2 by doubling its contention window each time until dropping it and is not responded with CTS packet by node 2. As soon as node 4

finishes the transmission, it resets its contention window and hence has got higher priority than other nodes for occupying the channel for next transmission. Again, node 4 can start a new transmission to node 5 when node 1 is trying to establish connection with node 2, because it senses the channel as idle. Thus due to a greedy flow passing through the link 4 to 5, the flow 1 to 2 will be starved.

Now, consider the scenario of Figure 7, where we have five TCP flows. Flow 1 passing through the link 1 to 5 and flows 2, 3, 4 and 5 are passing through the links 2 to 4. If node 2 is transmitting the packets of flows 2, 3, 4 and 5 to node 3, then node 1 will not be able to get the idle channel until all the flows have finished their transmissions. Nodes 1 will continually doubling its contention window till maximum limit, CWmax, is reached. Furthermore, after finishing the transmission, node 2 will reset its contention window and thus gets priority over node 1 and again gets the free channel for next transmission. Thus flow 1 will be denied of free channel and will be starved.

THE propoSEd SoluTIon

The above stated problem of starvation of TCP flow 1 of Figure 7 can be solved by spatial reuse of the channel. It is known that spatial reuse of the channel can be achieved by scheduling simultaneous transmissions whose regions are not interfering to each other. Further, different flows may have contention of different levels. So flow with longer path length in terms of number of hops, in this case flow 1, suffers with more

Figure6.starvationofflow1duetogreedyTCPflow2



contention and requires allocating the channel resources for a long time share. This reduces the channel reuse and hence the average through-put. Again, since there is limitation of at least four hops distance between two simultaneous transmitting nodes, we can achieve concurrent transmissions within the interfering regions of four hops distance between two simultaneously transmitting nodes by using CTMAC protocol at the MAC layer instead of IEEE 802.11 MAC protocol. All the flows/ transmissions of Figure 7 can be rearranged as in Figure 8 (a), (b) and (c).

In the scenario of Figure 8 (a), the flow 1 is passed through the path 1-6-7-8-9-5 instead of 1-2-3-4-5, where node 6 is in the transmis-sion range of nodes 1, 2 and 7 but node 1 and 7 are not sharing the same transmission regions. Similarly node 7 is in the transmission regions of nodes 6, 2, 3 and 8, whereas node 8 is not sharing the same transmission region with node 2 and node 6 is also not in the transmis-sion region of node 3, and so forth for nodes 8 and 9. Again in Figure 8 (b) and (c) flow 1 is achieved in two time slots T1 and T2 of equal duration, because in CTMAC protocol, the secondary transmission cannot transmit after primary transmission has finished due to col-lisions between DATA packets of secondary transmission and ACK packets of the primary transmission as shown in Figure 5. In T1 time slot concurrent transmissions 1-6, 7-8, 9-5 and 2-3 are achieved. In the next T2 time slot concur-

rent transmissions 3-4, 6-7, 8-9 are achieved. Within each time slot, the transmission of all the DATA and ACK packets is scheduled, such that there is no interference between the ACK packets of the previous time slot and the DATA packets of the current time slot.

SIMulATIon rESulTS

We first consider the TCP throughput of different flows in the scenario of Figure 7. As shown in the Figure 9, the throughput of flow 1 is starved due to greedy nature of flows 2, 3, 4, and 5 as TCP favors short flows; specially the one or two hop flows and penalizes long flows. For the one or two hop flows, since each node along the path can sense other node’s transmission, there is no hidden node within the path. If there is no competing one or two-hop flows in the vicinity, they try to occupy all the available bandwidth and thus showing high throughput. Flow 1, as a victim, encounters sever medium contention from all other flows and gain no throughput at all. Other four two-hop flows compete with each other along the same path and approximately fairly share the bandwidth with a slight variation. Scheduling of concur-rent transmission can solve the congestion and starvation of TCP flows by utilizing the capacity of those nodes which are in the transmission range of the sender or receiver, which were restricted from transmission, when IEEE 802.11

Figure7.StarvationofTCPflow1



was used, due to NAV information in the RTS/ CTS packets of the scheduled sender/receiver. We use ns-2.31 (http://www.isi.edu/nsnam/ns) to conduct the simulations. The transmission range is about 250m and the sensing range is about 550m. We set the channel bandwidth as 2Mbps and use 1000 bytes as the payload size of each data packet.

We have compared the performance of the concurrent transmission protocol (CTMAC) for MANET, in Figure 10, having 50 mobile nodes and flows of 4-hop distance between transmitter and receiver, and SANET 5-node chain topol-ogy. The simulation time is 200 seconds, speed of mobile nodes in MANET is 10 meters per

second, having random topology. As we can see in the figure that the throughput of the SANET is much higher at nearly 550 Kbps and is stable at that value due to fixed node positions, but for MANET the end-to-end connectivity depends on the location and speed of the intermediate nodes and thus after gaining a value of 480 Kbps approximately it starts deteriorating and comes down. As the intermediate nodes move dynamically in MANETs, such that end-to-end connectivity cannot be maintained and the total throughput may come down to a lower value. This movement of nodes makes the throughput unstable in MANETs.

Figure8.(a),(b),(c):Rearrangementofflow1intopologyscenarioofFigure7



We consider the chain topology of Figure 7 and scheduling the concurrent transmissions (1-2, 3-4) and (2-3, 4-5) using CTMAC protocol in two time slots of equal duration each of (Tij

Ack-Tij

Data), as shown in Figure 5. The same topology is simulated using IEEE 802.11 MAC protocol and compared the aggregate throughput

of all five TCP flows with the CTMAC protocol aggregate throughput. The simulation time is 300 seconds. We observe in Figure 11 that the throughput of the IEEE 802.11 MAC protocol is nearly 750 Kbps as compared to 2.6 Mbps for CTMAC protocol. This huge difference is due to the fact that when we are using IEEE 802.11

Figure9.PerformanceofscenarioofFigure7usingIEEE802.11MAC

Figure 10. Performance comparison of CTMAC for MANET and SANET 5-node chaintopology



MAC protocol; there can be only one transmis-sion possible within this four hop distance as compared to two simultaneous transmissions in the case of CTMAC protocol. Again, in the case of IEEE 802.11 MAC, flow 1 is completely starved and the aggregate throughput is due to remaining four flows only as compared to the CTMAC. So by enabling multiple concurrent transmissions we can improve the aggregate throughput of the network.

The simulation results for the scenario of Figure 8 (b) and (c) are presented in Figure 12 where we observe that as the flow 1 has different path, 1-6-7-8-9-5, it has full available bandwidth as compared to other flows that have the shared network resources. So the throughput of flow 1 with dedicated path and scheduled concurrent transmissions is much higher, nearly 2.6 Mbps, as compared to other flows having individual throughput of approximately 0.23 Mbps. Again if we are using IEEE 802.11 MAC protocol, we observe that the flow 1 is starved and other flows are having individual throughput of nearly 0.17 Mbps. This increase of 0.06 Mbps for each individual throughputs of flow 2, 3, 4, and 5 in the case of CTMAC protocol as compared to IEEE 802.11 MAC protocol is due to the fact that after the flow 1 has been scheduled through

a different path link, the remaining flows share the additional bandwidth which was consumed earlier by high medium contentions.

In Figure 13, the simulation results demon-strate that CTMAC always achieve significantly shorter end-to-end delay than IEEE 802.11 MAC does for the scenario of Figure 8 (b) and (c). We also observe that in CTMAC, the end-to-end delay is proportional to the flow path length in terms of number of hops.

ConCluSIon

Congestion control is decisive to the reliable transport level service in wireless ad hoc net-works. Starvation is a severe unfairness problem due to medium contention suffered by TCP flows, which contribute to the severe network congestion. IEEE 802.11 MAC protocol uses DCF mode of operation which restricts any simultaneous transmission in the region of conflict. By using concurrent transmission protocols, like CTMAC, we can schedule multiple simultaneous transmissions within the interfering range of a node. With extensive simulations we have shown that starved TCP flow with longer path can be allocated with a dedicated path within the conflicting regions

Figure11.Performance comparisonofCTMACand IEEE802.11MAC for the scenarioofFigure 7



Figure12.Flow-wiseperformancecomparisonofCTAMCandIEEE802.11

Figure13.AggregateEnd-to-EnddelaycomparisonofIEEE802.11andCTMAC



thereby increasing the spatial reuse of the chan-nel and hence the aggregate throughput. The results show that combination of traditional TCP and IEEE 802.11 MAC performs poorly in wireless ad hoc networks in comparison to the combination of traditional TCP and concurrent transmission protocol, CTMAC, resulting in higher aggregate throughput and lower end-to-end delay and also solves the starvation problem of TCP flows with long paths.

rEFErEnCES

Acharya, A., Bansal, S., & Misra, A. (2003). MACA-P: A MAC for concurrent transmissions in multi-hop wireless networks. In . Proceedings of IEEEPerCom, 2003, 505–508.

Acharya, A., Bansal, S., & Misra, A. (2004). Design and analysis of a cooperative medium access scheme for wireless mesh networks. In . Proceedings of BroadNets, 2004, 25–29.

Bagrodia, R., Martin, J., Ren, A., & Takai, M. (2002, June). Directional virtual carrier sensing for directional antennas in mobile ad hoc networks. Paper presented at ACM MobiHoc.

Balanis, C., et al. (2000). Smart antennas for future reconfigurable wireless communication networks. Tempe, AZ: Arizona State University. Retrieved from http://www.eas.asu.edu/trc-comm/nsf/presentations/Annual_report.pdf

Bandyopadhyay, S., Hausike, K., Horisawa, S., & Tawara, S. (2001, October). An adaptive MAC and directional routing protocol for ad hoc wireless networksusingESPARantenna. Paper presented at ACM MobiHoc.

Bharghavan, V. (1998, August). Performance analysis of a medium access protocol for wireless packet networks. Paper presented at the IEEE Performance and Dependability Symposium, Raleigh, NC.

Boppana, S., & Shea, J. M. (2009). Overlapped carrier-sense multiple access (OCSMA) in wireless ad hoc networks. IEEE Transactions on MobileComputing, 8(3). doi:10.1109/TMC.2008.114

Cao, J., Chan, A. T. S., Chan, K. C. C., Leong, H. V., Wang, X., & Yu, W. (2008). A high-throughput MAC protocol for wireless ad-hoc networks. IEEETransactions on Wireless Communications, 7(1), 135–145. doi:10.1109/TWC.2008.06094

Chen, K., Nahrstedt, K., & Xue, Y. (2003). On setting TCP’s congestion window limit in mobile ad hoc networks. ProceedingsoftheIEEE, ICC, 2003.

Chen, X., Fang, Y., Zhai, H., & Wang, J. (2004). TCP performance over mobile ad hoc networks. Canadian Journal of Electrical and ComputerEngineering, 29(1/2), 129–134. doi:10.1109/CJECE.2004.1425806

Choudhury, R. R., Ramanathan, R., Vaidya, N. H., & Yang, X. (2002, September 23-24). Using directional antennas for medium access contol in ad hoc networks. Paper presented at ACM MOBICOM.

Crompton, R., & Ward, J. (1992). Improving the performance of slotted ALOHA packet radio networks with adaptive arrays. IEEETransactions on Communications, 40(11), 1716–1724. doi:10.1109/26.179935

Fu, Z., Gerla, M., Luo, H., Lu, S., Zerfos, P., & Zhang, L. (2003). The impact of multihop wireless channel on TCP throughput and loss. In Proceedings ofIEEEINFOCOM2003.

Fu, Z., Lu, S., & Meng, X. (2002). How bad TCP can perform in mobile ad-hoc networks. In Proceedings of 2002 IEEE Symposium on Computers andCommunications.

Gerla, M., Qi, L., Shu, Y., & Xu, K. (2003). Enhancing TCP fairness in ad hoc wireless networks using neighborhood RED. In Proceedings of ACM MobiCom 2003.

Giles, T., Sanchez, M., & Zander, J. (2001, March 5-6). CSMA/CA with beam forming antennas in multi-hop packet radio. Paper presented at the Swedish Workshop on Wireless Ad Hoc Networks.

Gupta, P., & Kumar, P. R. (2000). The capacity of wireless networks. IEEE Transactionson Information Theory, IT-46(2), 388–404. doi:10.1109/18.825799

Hiromoto, R., Nasipuri, A., Ye, S., & You, J. (2000, September). A MAC protocol for mobile ad hoc networks using directional antennas. Paper presented at IEEE WCNC, Chicago.

IEEE. (1999). International Standard ISO/IEC802.11; Part 11: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY)specifications (ANSI/IEEE Std. 802.11). Washington, DC: IEEE Computer Society LAN/MAN Standards Committee Information Sciences Institute. (n.d.). Thenetworksimulatorns-2. Retrieved from http://www.isi.edu/nsnam/ns



Kim, J. H., & Lee, J. K. (1999). Capture effects of wireless CSMA/CA protocols in Rayleigh and shadow fading channels. IEEE Transactionson Vehicular Technology, 48(4), 1277–1286. doi:10.1109/25.775376

Ko, Y. B., Shankarkumar, V., & Vaidya, N. H. (2000, March). Medium access control protocols using directional antennas in ad hoc networks. Paper presented at IEEE INFOCOM.

Krunz, M., & Muqattash, A. (2005). POWMAC: A single-channel power control protocol for throughput enhancement in wireless ad hoc networks. IEEEJournal on Selected Areas in Communications, 23(5), 1067–1084. doi:10.1109/JSAC.2005.845422

Lee, S. J., Perkins, C., & Royer, E. (2000, September). The effects ofMACprotocols onadhocnetworkcommunication. Paper presented at IEEE WCNC, Chicago.

Zander, J. (1990). Slotted ALOHA multi-hop packet radio networks with directional antennas. ElectronicsLetters, 26(25). doi:10.1049/el:19901351

RajeshVermareceivedtheBEdegreeinelectronicsandcommunicationengineeringandMEdegreeindigitalsystemsfromtheDepartmentofElectronicsandCommunicationEngineering,MotilalNehruNational Institute ofTechnology, India in 1994and2001 respectively.He iscurrentlyworkingtowardsthePhDdegreeattheDepartmentofElectronicsandCommunicationEngineering,MotilalNehruNationalInstituteofTechnology,India.Hisresearchinterestsarein the area of multi-hop wireless ad hoc networks, cross layer optimizations, MAC protocols, congestion control, smart antennas.

ArunPrakashreceivedtheBEdegreefromtheDepartmentofElectronicsandcommunicationEngineering, MJP Rohilkhand University, India in 2001, and the MTech degree in digitalcommunicationfromUttarPradeshTechnicalUniversity,Indiain2006.HeiscurrentlyworkingtowardsthePhDdegreeat theDepartmentofElectronicsandCommunicationEngineering,MotilalNehruNationalInstituteofTechnology,India.Heunderwentsixmonthsresearchandgraduate student training at University of Waterloo, Canada during September 2008-February 2009. His research interests are in the area of wireless and mobile networks, with emphasis on mobility management of mobile networks.

RajeevTripathireceivedhisBTech,MTech,andPhDdegreesinelectronicsandcommunicationengineering form Allahabad University, India. At present, he is a professor in the Department of ElectronicsandCommunicationEngineering,atMotilalNehruNationalInstituteofTechnology,Allahabad, India.Heworkedasa facultymemberat theUniversityofTheWest Indies, St.Augustine,Trinidad,WI,duringSeptember2002-June2004.HewasavisitingfacultyatSchoolofEngineering,LiverpoolJohnMoorseUniversity,UK,duringMay-June1998andNov-Dec1999. He was also involved in a research project on “Performance analysis and development ofATMSwitcharchitectureforB-ISDN”,jointlysponsoredbygovernmentofIndiaandBritishgovernment, under Indo-UK science and technology research fund. He worked as reviewer of IEEE Communication Letters and West Indian Journal of Engineering. He served as program co-chair of the First International Conference on Computational Intelligence, Communication Systems, and Networks, held in Indore, in July 2009. He is on the program committee of several international conferences in the area of wireless communication and networking. His research interestsareHighspeedcommunicationnetworks,Performanceofnextgenerationnetworks:



switching aspects, MAC protocols, mobile ad hoc networking, and mobility management of vehicular networks.

NeerajTyagireceivedthePhDdegreeinhybridmodelwithmodifiedMANETandMIProutingforInternetaccessfromMNNIT,AllahabadDeemedUniversityinyear2008.HejoinedMNNIT(FormerlyMNREC)Allahabadaslecturerintheyear1989andhasbeenworkingasanassistantprofessoratMNNITintheDepartmentofComputerScienceandEngineeringfromSeptember2000.Hehasabout19yearsofteachingexperience.HeobtainedhisMTechincomputerscienceandengineeringfromAllahabadUniversityin1997anddidhisBTechincomputersciencefromAllahabad University India in 1987. His areas of interest are computer network, routing in wirelessnetwork,MobileIP,MANET,networkadministrationandUNIXsystemadministration.HealsoworkedforWarmanInternational,Australia,GECapital,USAandElectronicsDataSystems, USA as UNIX and network system administrator through Birla Soft India for two years from 1999 to 2001.

Improving Throughput of Starved TCP Flow by Sidestepping Bottleneck Nodes Using Concurrent...

Documents

Transcript of Improving Throughput of Starved TCP Flow by Sidestepping Bottleneck Nodes Using Concurrent...