Cross-layer TCP with bitmap error recovery scheme in ... - CiteSeerX

Telecommun Syst DOI 10.1007/s11235-009-9226-1

Cross-layer TCP with bitmap error recovery scheme in wireless ad hoc networks Rung-Shiang Cheng · Der-Jiunn Deng · Yueh-Min Huang · Lianfen Huang · Han-Chieh Chao

© Springer Science+Business Media, LLC 2009

Abstract Current TCP is not able to distinguish corruption losses from packet loss events. Hence, high transmission errors and varying inherent latency within a wireless network would cause seriously adverse effects to TCP performance. To improve TCP in IEEE 802.11 multi-hop ad hoc wireless networks, this study proposes an error recovery mechanism based on coordination of TCP and IEEE 802.11 MAC protocols. The simulation results confirm that the proposed error recovery approach could provide a more efficient solution for frequent transmission losses, and enable TCP to

R.-S. Cheng () Department of Computer and Communication, Kun Shan University, No. 949, Dawan Rd., Yongkang City, Tainan County 710, Taiwan, ROC e-mail: [email protected] D.-J. Deng Department of Computer Science and Information Engineering, National Changhua University of Education, Bao-Shan Campus, No. 2, Shi-Da Rd., Changhua City 500, Taiwan, ROC e-mail: [email protected] Y.-M. Huang Department of Engineering Science, National Cheng Kung University, No. 1, University Rd., Tainan City 701, Taiwan, ROC e-mail: [email protected] L. Huang Department of Communications Engineering, Xiamen University, No. 422, Simingnan Rd., Xiamen City, Fujian Province, 361005, China e-mail: [email protected] H.-C. Chao Department of Electronic Engineering, National Ilan University, No. 1, Sec. 1, Shen-Lung Rd., I-Lan City 260, Taiwan, ROC e-mail: [email protected]

distinguish between congestion errors and transmission errors, and thus, to respond with proper remedial actions. Keywords Multi-hop · Cross-layer design · Bitmap · Ad hoc · IEEE 802.11

1 Introduction Wireless technologies provide mobile access to networks and eliminate the need of fixed cable infrastructures, thus enabling cost-effective network deployment. In recent years, wireless communication networks have been extensively deployed and are generally specified in accordance with IEEE 802.11 standards. However, a wireless link is commonly characterized by an unpredictable bit-error rate and varying latencies [1], hence, wireless environments pose great challenges when attempting to provide reliable data transmission for transport protocols, such as TCP. TCP is the most widely applied transport layer protocol for achieving reliable data transfer services over the Internet. As wireless networking gains in popularity, TCP continues to be the leading transport layer protocol. TCP guarantees end-to-end error-free data delivery services, and was originally designed for wired networks with a relatively reliable physical layer, in which packet losses arise primarily because of network congestion. When running TCP over wireless networks, however, packet losses due to network congestion rarely occur, but result primarily from repeated contentions or erratic errors in wireless links. The current TCP format is not able to distinguish between transmission errors and network congestion. Once packet losses are detected, TCP considers the event as an indication of network congestion and invokes a congestion control mechanism that results in an undesirable reduction

R.-S. Cheng et al.

of transmission rate. To improve TCP performance in wireless networks, it is important to handle packet losses caused by network congestions and wireless transmission errors differently. This paper investigates difficulties arising when utilizing the TCP protocol in wireless networks. The performance of data transfer relies on support from the end-to-end transport layer, as well as on the quality of the hop-by-hop communication link. Therefore, the present study attempts to enhance the legacy of IEEE 802.11 MAC and TCP protocols, in such a manner that TCP is rendered capable of differentiating between congestion and corruption losses, using information received from the MAC layer. The remainder of this paper is organized as follows. Section 2 provides a brief description on the 802.11 standard. Section 3 describes major challenges arising when utilizing TCP in multi-hop wireless networks. Section 4 presents the proposed Bitmap-based error recovery scheme based on a cross-layer design. Section 5 discusses the results of evaluation simulation performances. Finally, Sect. 6 presents brief conclusions.

2 Preliminaries The IEEE 802.11 standard [2] defines two types of services, namely a contention-free polling-based point coordination function (PCF) and a contention-based distributed coordination function (DCF). PCF is a centralized scheme, while DCF is distributed. In infrastructure-based networks, DCF can operate alone or in conjunction with PCF, however, in ad hoc networks, DCF operates alone. DCF is the basic access method for the 802.11 standard and is based on the conventional carrier sensing multiple access with a collision avoidance (CSMA/CA) scheme. DCF comprises both a basic access method and an optional channel access method based on RTS/CTS exchanges. In 802.11, priority access to the wireless medium is controlled by the application of an inter-frame space (IFS) time between the frame transmissions. To prevent collisions, the transmitter is obliged to wait for the channel to remain free Fig. 1 Basic access mechanism in DCF

Fig. 2 Virtual carrier sensing mechanism in DCF

for a specified interval of time, designated as the distributed inter-frame space (DIFS), before sending a frame. If the medium is currently busy, or becomes busy during this interval, the transmitter defers the frame transmission until it detects a DIFS. At this point, the transmitter selects a random interval, referred to as the backoff time, to determine the moment to commence transmission. The backoff time is an integer number of slots, uniformly chosen from the interval (0, CW-1), where CW is the backoff window, also referred to as the contention window. The backoff number counts down slot-by-slot, and when it reaches zero, the frame is transmitted. Due to the half-duplex nature of wireless interfaces, stations in the network are unable to detect a collision simply by listening to their own transmissions. Therefore, an immediate positive acknowledgment technique is employed to confirm the successful reception of a frame. Specifically, having received a frame, the receiver waits for a short interframe space (SIFS), and then transmits a positive MAC acknowledgment to the transmitter, confirming that the frame has been correctly received. The SIFS is deliberately assigned a shorter interval than the DIFS in order to assign the receiving station a higher priority than any other stations waiting to make a transmission. The ACK is only transmitted if the frame is received correctly, and hence if the transmitter does not receive an ACK, it assumes that the data frame must have been lost, and therefore, schedules a retransmission. Figure 1 illustrates the basic operations involved in 802.11 DCF. To alleviate the hidden-station problem [3], 802.11 DCF also provides an optional channel access method using a virtual carrier sensing mechanism based on two special control frames, namely request-to-send (RTS) and clear-to-send (CTS). As shown in Fig. 2, before transmitting a frame, the transmitter transmits an RTS frame asking the receiver if the medium in its vicinity is free. Once the receiver receives this RTS frame, and assuming no transmission interference is present, it waits for the specified SIFS interval, and then sends a CTS frame to the transmitter. Both transmitter and receiver neighbors overhear these frames and consider the medium reserved for the transmission duration.

Cross-layer TCP with bitmap error recovery scheme in wireless ad hoc networks

3 Transport layer challenges in wireless networks TCP is a reliable end-to-end acknowledgment-clocking window based protocol. TCP controls the sending rate using a congestion window parameter [4]. In theory, TCP should be independent of the technology used to implement the underlying infrastructure. However, in practice, the high error rates typical of wireless networks generally cause the backoff mechanism to be inappropriately invoked. Consequently, the utilization of the network bandwidth is severely degraded. Packet losses in a wired network are mainly caused by buffer overflows at the bottleneck router. However, in a multi-hop wireless network, packet losses due to buffer overflows at intermediate stations rarely occur (unless the station buffer is very small), but result primarily from linklayer contentions or transmission errors [5]. Collision occurrences in a shared channel increase as in-flight packet numbers increase, and hence, a large sized TCP window leads to a higher degree of link-layer contention, due to the halfduplex nature of wireless links, and thus to a higher number of dropped packets. As discussed in [6] and [7], channel access contentions may occur between different flows passing through the same vicinity, or between different packets within the same flow (e.g. consider the case where the forwarded TCP data competes for the channel with the backward ACK of the previous data), causing TCP transmission rate to fall as a result of frequent invocations of the congestion control mechanism. Moreover, frequent packet exchanges taking place in multi-hop wireless networks exacerbate channel contention problems, resulting in more packets dropped in the MAC layer. This exacerbation causes inappropriate invocation of the TCP back-off mechanism, further degrading network bandwidth utilization. Several recent studies [8–11] have proposed alleviating the effects of non congestion-related losses in TCP performance over networks with wireless links, by introducing intelligence at the base station using TCP-aware SNOOP mechanisms, or by splitting the entire path into two distinct parts, namely the wired connection and the wireless connection. Under this approach, upon successful transmission of a packet in one connection, an acknowledgement message is sent to the TCP source and the packet is then relayed to the next connection. However, a major disadvantage of this split-connection approach is that the base station is required to maintain information relating to the state of both connections, as well as caching unacknowledged packets for every TCP connection passing through it, causing an overflow problem. The Explicit Loss Notification (ELN) scheme [8] has been proposed as a means to provide TCP with the ability to differentiate between congestion and wireless losses.

However, because lost packets can only be retransmitted after the round-trip time has elapsed, error recovery is slow compared to that achieved by SNOOP mechanisms. The TCP-Feedback and TCP-ELFN [12] schemes aim to improve TCP performance by sending an Explicit Link Failure Notification (ELFN) from the node immediately upstream of the link failure to every TCP connection passing through that link. Although ELFN-based approaches perform far better than conventional TCP schemes in mobile scenarios, such approaches typically provide lower throughputs in static networks [13]. The end-to-end scheme SACK [14], uses a SACK option carried by the return selective ACK to inform the sender of successfully received data. However, the SACK option is not bit-efficient in a contention-based 802.11 DCF environment due to its use of two 32-bit sequence numbers to specify a single SACK block. The resulting limitation of three maximum SACK blocks per ACK segment (when other options, e.g. Timestamp, are also used), is too restrictive for TCP over erratic wireless links [15]. In [16], Mascolo et al. introduced the TCP Westwood scheme. This scheme estimates available bandwidth based on the measured inter-arrival rate of successive ACK packets, and then uses this estimate to set appropriate values of the slow-start threshold and the congestion window size. This approach avoids the default blind-halving of the window size, as applied in conventional TCP (e.g., Reno), and enables TCP Westwood to achieve a high link utilization in the presence of random sporadic losses, typical of wireless links. However, TCP Westwood generally overestimates the available bandwidth because of ACK packet clustering. As shown in [17], in a heterogeneous scenario, TCP Westwood consistently achieves a higher throughput than its fair share. In other words, Westwood improves the performance of the TCP connection, but inevitably introduces a trade-off between its throughput and its friendliness to other TCP implementations.

4 Cross-layer TCP with bitmap error recovery scheme Reliable data transmission requires that the standard TCP protocol employs an ACK mechanism to confirm successful packet reception. Although the transport protocol is errortolerant, wireless losses impose additional error recovery requirements. Efficient error recovery is therefore vital for maintaining acceptable application performance. 4.1 Proposed bitmap-based error recovery procedure The present study applies the selective negative acknowledgement (SNACK) option to improve TCP performance in wireless networks. SNACK is similar to the solution

R.-S. Cheng et al. Fig. 3 SNACK bitmap example

standardized as cumulative ACK with bitmap in IEEE 802.16 [18]. The basic SNACK retransmission mechanism remains in TCP ACK, however, packet loss information is constructed with the bitmaps [19], to reduce acknowledgment overhead. Whenever lost packets indicated by SNACK are retransmitted in the proposed bitmap-based error recovery mechanism, the sender sets the expected recovery sequence number to the highest sequence number sent so far in order to keep track of the recovery segment in the error recovery phase. The “error recovery procedure” (which includes fast retransmission and fast recovery [4]) is triggered when the sender receives a SNACK, and terminates when it receives an acknowledgement beyond the expected recovery sequence number. The sender reconfigures its congestion window and slow-start threshold when the third duplicate ACK is received, and then enters into the fast recovery phase. Let W and Wt denote the current congestion window size and slow-start threshold, respectively. The TCP sender recovers the lost segments based on the ACK received from the TCP receiver, in accordance with the following functions: Case 1. If receipt of a new ACK with a sequence number larger than, or equal to, the expected recovery sequence number: Resume Congestion Avoidance phase. Case 2. If receipt of duplicate ACK (carrying SNACK option): Increment duplicate ACK count for segment being ACKed. Fast Retransmit phase: Set Wt = W/2 and then set W = Wt . Fast Recovery phase: If the ACKed bitmap length is larger than previous bitmap length: If there is an additional hole(s), retransmit the missing data in the corresponding MSS-sized block(s). If the ACKed bitmap length equals previous bitmap length: Retransmit the first hole indicated in the SNACK option.

If the ACKed sequence number beyond the expected recovery sequence number cached in the receiver buffer (identified by SNACK option): Retransmit next expected segment. Case 3. If Receipt of partial ACK (carrying SNACK option): If the partial ACK number is beyond the end of the sequence space, as indicated by the previous SNACK option, retransmit any segments necessary to fill the holes. Else: Retransmit the first hole indicated in the SNACK option. Case 4. Upon timer expiry: Set Wt = W/2, then set W = 1; Recover the “missing segments” from the SlowStart phase. During the fast recovery phase, the sender increases congestion window size by one segment each time it receives a duplicate ACK (piggyback SNACK option). The TCP sender transmits a new segment to fill the pipe between the source and the destination, if possible, by the new window size. The sender identifies any holes in the receiver buffer by inspecting the SNACK option, therefore verifying whether the retransmitted segments have been successfully received (see case 2(a) and 2(b)). If the SNACK option indicates that the segment beyond the expected recovery sequence number has been cached in the receiver buffer (see case 2(c)), the sender retransmits the first hole indicated in the SNACK option and updates the retransmission timer, because the previously retransmitted packet may have been lost again. As illustrated in Fig. 3(a), by examining the ACKed sequence number and the historical information provided by the SNACK option, the sender determines that retransmitted segment 1 has not yet been received, and must therefore be retransmitted. If the returned SNACK for the previous retransmitted packet acknowledges a new ACK sequence number (see case 3, as illustrated in Fig. 3(b)), this ACK is deemed a partial ACK because not all transmitted packets have been acknowledged. The sender in this event remains in the fast recovery phase until receiving an acknowledgement beyond the expected recovery sequence number. The error recovery procedures described above alleviate the multiple-packet-loss problem in TCP, i.e. retransmission

Cross-layer TCP with bitmap error recovery scheme in wireless ad hoc networks

is not governed by timeouts as in conventional error recovery schemes. When retransmitted packets are repeatedly lost because of transmission errors, these packets can be retransmitted immediately upon SNACK receipt. This ability to respond robustly to errors is particularly important in wireless environments, characterized by relatively high corruption probability of both packets and their associated ACKs. Since TCP data packets and their ACKs contend for the same channel, reducing ACK flow frees up more bandwidth for the TCP data packets. The number of transport level ACKs should be reduced to alleviate TCP self-collision problems [20], allowing TCP to work more efficiently in 802.11 wireless environments. The SNACK receiver in the current implementation is extended to support a delayed ACK scheme. A delayed ACK in this scheme is generated for every n TCP packets, or after a specified time interval, t, if remnant n-1 packets have not yet arrived. In the standard delayed ACK option, n is assigned a value of 2 [21]. 4.2 Extension to IEEE 802.11 MAC The link layer generally defines procedures for achieving reliable operations over a communication link. This study extends the IEEE 802.11 DCF scheme by introducing a crosslayer modification to reduce link layer contention effects on TCP performance. In the proposed approach, whenever a frame transmission failure occurs, the extended MAC layer protocol retransmits the corrupted packet in accordance with the following functions: Case 1. If retry limit parameter > retry threshold (i.e. seven for basic access mechanisms and four for virtual carrier sensing mechanisms): If the number of retransmission attempts (RT ) < retransmission threshold as well as receipt of TCP ACK within the same flow: Forward the associated cross-layer negative acknowledgement (NACK) of the dropped packet piggybacks using the reverse TCP ACK along the end-to-end path; Increase the value of RT . Else: Discard the transmitted packet and then reset the contention window (CW) and retry limit parameter, respectively. Case 2. Otherwise: Increase CW and launch the backoff procedure Increase the value of retry limit (specified in the 802.11 standard). Specifically, a NACK is triggered when a link layer frame is dropped because of continuous transmission failures. If a

TCP data frame is discarded after several retransmission attempts, the 802.11 MAC protocol triggers an explicit corrupted notification associated with the sequence number of the dropped TCP packet, and then piggybacks the NACK option using a reverse ACK to notify the TCP sender to retransmit the missing packet. Hence, by inspecting the ACKed sequence number and historical NACK information, the TCP sender clearly identifies corruption losses in the wireless links, and therefore, determines whether invoking congestion control mechanisms is appropriate. Note that the NACK notification is sent piggyback mode, with the return TCP ACK segment, in order to avoid increasing link layer contention. This cross-layer support from the MAC layer protocol ensures that the transport layer protocol is aware of transmission error in the link layer and can then react to this error in accordance with wireless corruption information conveyed by the received NACK.

5 Performance analytic and simulation results This study investigates TCP performance obtained in a static multi-hop ad hoc environment using the ns-2 network simulator [22]. System configuration and simulation results are described in the following. 5.1 Error model Empirical observations show that the loss characteristics of a wireless channel are bursty due to various fading effects [23, 24]. The present simulations apply an extended Gilbert–Elliott (GE) [25] error module to the 802.11 wireless link in the ns-2 network simulator to reflect the bursty effects of wireless transmission errors, and to mimic fading in the communication channel. The rate at which errors occur in the GE model is dependent on channel conditions. Fig. 4 presents the state diagram of the Gilbert–Elliott error model. In a good state, G, losses occur with a low probability, PG , whereas in a bad state, B, the channel operates in fading condition, and loss probability, PB , is correspondingly higher. The steady state probabilities of being in states G and B are given by πG = PBG /(PGB + PBG ) and πB = PGB /(PGB + PBG ), respectively. The average packet loss rate produced by the GE model is P = PG πG + PB πB .

Fig. 4 State transition diagram for the Gilbert-Elliott model

R.-S. Cheng et al. Table 1 Parameters for the MAC and PHY Layer

5.2 Physical and data link layer model

802.11

Simulations performed in this study use a two-ray ground reflection model [26] to model signal propagation in a wireless network. In the two-ray ground reflection model, the received power, Pr , at a distance d from the transmitter is expressed as: h2t h2γ Pγ = Pt Gt Gγ 4 d

(1)

where Gt is the transmitter antenna gain, Gr is the receiver antenna gain, d is the distance between the antennas in meters, ht is the height of the transmitter, and hr is the height of the receiver. The default transmitting power Pt in ns-2 is 0.28, and hence the transmission range and physical carrier sensing range are calculated as 250 meters and 550 meters, respectively. However, as discussed in [27], default values of transmission range and carrier sensing range used in ns-2 are two to three times higher than the values measured in practice. To accurately model the attenuation of the communication radius between antennas close to the ground, the current simulation model specifies transmission range as 40 meters and carrier sensing range, with respect to the transmitting station as 85 meters. This work assumes that receiver and transmitter antennas have a height of 1.5 meters and operate in the 2.4 GHz band. Table 1 summarizes parameter values used in the simulations for the different 802.11 standard specifications. The RTS, CTS, and ACK packets are assigned default sizes of 20 bytes, 14 bytes, and 14 bytes (excluding PHY hdr ), respectively. Note that PHY hdr indicates the sum of the PLCP preamble and the PLCP header, and DIFS = 2 · SLOT time + SIFS. 5.3 One-hop TCP throughput The Sect. 2 discussion makes it is clear that the overall time required to successfully transmit a frame is given by: DIFS + TDATA + SIFS + TACK + τ , where TDATA is the time required to transmit a DATA frame (PHY hdr + MAChdr + MAC_Payload + FCS), TACK is the time required to transmit a ACK frame, and τ is the average backoff time. Let m denote data length at the application layer. The expected throughput for a TCP connection over an 802.11 wireless link under the basic access scheme is given by: ThnoRTS/CTS =

m 2 · (DIFS + SIFS + TACK + τ ) + TTCPDATA + TTCPACK (2)

SLOT

802.11b

802.11g

50 µsec

20 µsec

9 µsec

SIFS

28 µsec

10 µsec

10 µsec

DIFS

128 µsec

50 µsec

PHY hdr

128 bits

192 bits

28 µsec 192 bits

CW min

32

32

32

CW max

1024

1024

1024

where TTCPDATA is the time required to deliver a MAC frame carrying the TCP data packet and TTCPACK is the time required to deliver a MAC frame carrying the TCP ACK. If the frame exchange is carried out using the RTS/CTS access method, the overheads associated with RTS and CTS control frame transmission must be added to the (2) denominator. Hence, the expected TCP throughput, ThRTS/CTS , is given by: ThRTS/CTS m = 2·(DIFS+3·SIFS+T +T +T RTS CTS ACK +τ )+TTCPDATA +TTCPACK (3)

where the value of DIFS + 3 · SIFS + TRTS + TCTS + TACK represents the expected MAC layer overhead induced by 802.11 DCF (RTS/CTS/ACK). Applying (2) and (3), Table 2 summarizes the expected one-hop TCP throughput with 1440-byte segments (MAC_Payload (1500)-TCP/IPHeader (40)-TCPOption (20) = 1440 bytes) over 802.11 wireless networks, specified in accordance with the settings indicated in Table 1. The average backoff time, τ , is set to (CW min /2) · SLOT to simplify throughput calculation. The results presented in Table 2 show that the 802.11 preamble and header introduce a considerable overhead, particularly when TCP runs over a higher-speed wireless link. The results also reveal that the RTS/CTS/ACK exchange has significant influence on end-to-end performance. Furthermore, the virtual carrier sensing mechanism apparently fails to improve performance relative to that provided by the basic access mechanism. This failure arises because the RTS/CTS frames are transmitted using the same signal channel used for the data frames, hence introducing an additional overhead. Table 3 summarizes the actual throughput (i.e. goodput) of a one-hop Reno TCP connection. Frame loss rate in a relatively clean wireless medium is typically less than 0.025 when using maximum size frames (i.e. PG < 0.025), as discussed in [3] and [24]. Hence, frame error rates (FERs), PG and PB , are assigned values of 0.0155 and 0.25, respectively, unless specified otherwise. Additionally, PGG , PGB ,

Cross-layer TCP with bitmap error recovery scheme in wireless ad hoc networks Table 2 Expected TCP throughput at different data rates

Type

Physical rate

802.11

2 Mbps

802.11b 802.11g

ThnoRTS/CTS /Achieved 1.49 Mbps/74.5%

1.36 Mbps/68.0%

11 Mbps

5.06 Mbps/46.0%

3.88 Mbps/35.2%

54 Mbps

14.98 Mbps/27.7%

7.86 Mbps/14.6%

Table 3 TCP Reno throughput

Table 5 TCP delayed SNACK throughput

ThnoRTS/CTS

ThRTS/CTS

1.13 Mbps

1.03 Mbps

802.11

802.11b

3.62 Mbps

2.46 Mbps

802.11g

11.44 Mbps

6.57 Mbps

802.11

ThRTS/CTS /Achieved

Table 4 TCP SNACK throughput

ThnoRTS/CTS

Improved

ThRTS/CTS

Improved

1.28 Mbps

13.27%

1.16 Mbps

12.62%

802.11b

4.37 Mbps

20.71%

3.05 Mbps

23.98%

802.11g

14.89 Mbps

30.15%

8.53 Mbps

29.83 %

ThnoRTS/CTS

Improved

ThRTS/CTS

Improved

option with delayed ACK. The performance enhancement is particularly evident in higher-speed links.

802.11

1.19 Mbps

5.30%

1.07 Mbps

3.88%

5.4 Multi-hop TCP throughput

802.11b

3.86 Mbps

6.62%

2.56 Mbps

4.06%

802.11g

12.75 Mbps

11.45 %

7.02 Mbps

6.84%

PBB , and PBG are specified as 0.94, 0.06, 0.86, and 0.14, respectively. This work assumes that frames may be corrupted because of channel noise or because they exhaust the retransmission threshold and are therefore discarded by the 802.11 MAC. The results presented in Table 3 show that the actual TCP throughputs are all lower than the expected maximum throughputs shown in Table 2. The results show that only a small percentage of the nominal bandwidth is actually used for data transmission. The higher-speed link is affected more significantly by the losses and backoff scheme, since TCP takes longer to reach its peak throughput following each loss. Although the one-hop ad hoc model considered here is not entirely representative of an actual ad hoc network, it nevertheless indicates the fundamental properties of TCP performance in wireless networks. Table 4 shows the throughput achieved by the SNACK scheme. The observed improvement in the TCP performance arises because of the more efficient treatment of frequent losses by SNACK than by the standard TCP scheme. Comparing the data in Table 4 with those in Table 3, it is observed that the performance of the higher-speed link is particularly improved. Although MAC layer collisions still waste bandwidth, SNACK improves TCP performance, since it avoids the requirement for slower timeout-based recovery routines, such as those employed in standard TCP. Table 5 shows SNACK throughputs obtained from the delayed ACK scheme (delayed packet number, n = 2, and delayed time interval, t = 0.1 sec). Comparisons of the data in Table 6 and Table 4 show a 29 percent improvement in the 802.11g links when the receiver employs the SNACK

This subsection examines the realizable capacity of the simple infrastructure-less ad-hoc wireless network, as shown in Fig. 5. In this configuration, the TCP sender is node 1, and the TCP destination is different according to the path length (hop count). The nodes are assumed to be static and separated from their immediate neighbors by a distance of 30 meters. Further, based on (1), the transmission range and physical carrier sensing range are configured at 40 meters and 85 meters, respectively. Table 6 shows variations of the average TCP throughput (with an introduced GE error) as a hop number function. The results show that the throughput is inversely proportional to hop distance. The TCP throughput in particular drops significantly as the number of hops increases first from 1- to 2-hops, and then to 3-hops. This phenomenon is due to carrier sense dependencies between wireless stations (the carrier sense range is 2.15 times that of the communication range in the simulation configuration). The SNACK delayed-ACK scheme provides slightly better performance than Reno, SACK, and Westwood schemes. The higher throughput results from a reduced number of selfcollisions in TCP (as a consequence of the delayed ACK scheme), and an improved response to multiple packet losses (as a consequence of the SNACK scheme). The cross-layer interaction between the transport and MAC layers has a critical effect on the detection of erratic errors and on the control of congestion. The results, presented in Table 7, show that the cross-layer (CS) design in 802.11 MAC provides the modified SNACK delayedACK(CS) scheme with more detailed information regarding the wireless link corruption error, compared to traditional TCP schemes. Therefore, the sender in the SNACK delayed-ACK(CS) scheme retransmits multiple lost packets

R.-S. Cheng et al. Table 6 TCP throughput with different path lengths Throughput (Mbps) TCP Flavor/Path length

1-hop

2-hop

3-hop

4-hop

5-hop

6-hop

7-hop 0.43

Reno

11.69

6.14

2.55

1.32

0.76

0.66

Reno delayed-ACK

13.88

6.90

3.14

1.93

1.18

0.86

0.63

SACK delayed-ACK

14.05

7.17

3.30

2.07

1.33

1.08

0.71

Westwood delayed-ACK

14.18

7.26

3.41

2.17

1.60

1.30

0.74

SNACK delayed-ACK

14.87

7.42

3.47

2.32

1.77

1.33

0.81

SNACK delayed-ACK(CS)

14.96

7.54

4.39

3.03

2.62

2.24

2.02

Fig. 5 Multi-hop network topology 1

Fig. 6 Multi-hop network topology 2

simultaneously, hence significantly reducing retransmission timeout occurrences. 5.5 Multi-hop TCP throughput versus loss probability This subsection performs simulations using the network topology, illustrated in Fig. 6, consisting of one TCP connection and one UDP connection (background interference traffic), respectively, to investigate the effects of wireless transmission errors and interference on TCP performance. The TCP connection in this figure extends from node 1 to node 4, while the UDP connection extends simply from node 5 to node 6. The background UDP transmission rate in the simulations is 64 Kbps with a constant bit rate, the frame error rate, PG , is set to 0.0155, while PB varies from 0.1 to 0.5. Meanwhile, the state probabilities PGG , PGB , PBB and PBG are set to 0.94, 0.06, 0.82 and 0.18, respectively. The results presented in Table 7 indicate that transmission error has influence on end-to-end performance, and the delayed ACK scheme alleviates the TCP self-collision

problem in wireless links, as discussed above. The SNACK delayed-ACK scheme (based on the proposed error recovery procedure) outperforms the SACK and Westwood schemes, because SNACK transmits lost packets immediately without having to wait for a timeout to occur, even for repeatedly lost packets. In environments prone to significant losses, application performance is affected not only by the rate at which TCP restores its transmission, but also by its own capability to recover from the transmission error. Although the Westwood scheme enables TCP to specify a more appropriate transmission rate in wireless environments, the error recovery routines inevitably degrade the congestion window size, resulting in undesirable reduction of the transmission rate, leading to significant performance degradation. The SNACK delayed-ACK(CS) scheme provides a better response to corruption packet losses than traditional TCP schemes (51% to 85% average throughput improvement, compared to the Reno delayed-ACK scheme, with different frame error rates), resulting from extended link layer protocol support. The scheme therefore eases the multiplepacket loss problem in wireless links, substantially reducing unnecessary reductions in window size. Hence, as shown in Table 7, the SNACK delayed-ACK(CS) scheme provides better performance than other TCP schemes, even in higher frame error rates and network congestion levels. Finally, to further examine the performance of the proposed mechanism under a complicated network environment, a series of simulations is conducted on randomly generated ad hoc network topologies. In this simulation, the network region is 100 m × 100 m with 20 wireless stations and TX range and PCSrange are set to 40 m and 85 m, respectively. At the beginning of each simulation, wireless stations are

Cross-layer TCP with bitmap error recovery scheme in wireless ad hoc networks Table 7 TCP throughput with different FER TCP Flavor/FER

Throughput (Mbps) 0.1

0.15

0.2

0.3

0.35

0.4

0.45

0.5

0.55

Reno

3.01

2.81

2.78

2.71

2.61

2.57

2.39

2.27

2.03

Reno delayed-ACK

3.02

2.88

2.81

2.79

2.63

2.54

2.50

2.27

2.19

SACK delayed-ACK

3.54

3.42

3.38

3.17

3.09

2.97

2.89

2.84

2.64

Westwood delayed-ACK

3.57

3.46

3.40

3.32

3.14

3.10

2.90

2.86

2.71

SNACK delayed-ACK

3.75

3.63

3.50

3.46

3.31

3.20

3.17

3.08

2.90

SNACK delayed-ACK(CS)

4.59

4.51

4.47

4.43

4.37

4.31

4.23

4.15

4.07

Table 8 TCP goodput comparison

References

Number of

Goodput (Mbps)

connection

Reno

Westwood

SNACK(CS)

2

11.56

11.82

14.72

3

9.71

10.62

12.62

4

8.75

8.78

10.94

5

7.24

8.40

9.44

6

6.52

7.32

8.95

7

6.50

7.16

8.48

8

6.14

6.51

7.68

9

5.92

6.24

7.45

randomly placed within this region and the number of shortlived TCP connections varies from two to nine. Moreover, the TCP source and the TCP destination of each connection are randomly selected. Table 8 presents the TCP goodputs of the proposed mechanism and other TCP flavors. As shown in Table 8, the TCP goodput is inversely proportional to the number of connections. This phenomenon is caused by competition for the bandwidth among TCP connections as the carrier senses dependencies over wireless stations. Comparing the results in Table 8, it is obvious that the proposed SNACK scheme achieves a better performance than those of other TCP schemes in multi-hop ad hoc wireless environments. 6 Conclusion The shared channel contention and erratic packet losses usually lead to a curbing of flow segments on the TCP, and thus limits the performance of TCP in wireless networks. This study developed a bitmap-based error recovery scheme to enhance TCP in IEEE 802.11 multi-hop wireless networks. An explicit negative acknowledgement scheme based on cross-layer design was further employed in both the TCP and 802.11 MAC to provide explicit corruption loss information. The simulation results show that the proposed mechanism could successfully alleviate the effects of radio channel errors on TCP performance, and thus greatly improve TCP in multi-hop ad hoc wireless environments.

1. Sundaresan, K., Hsieh, H. Y., & Sivakumar, R. (2004). IEEE 802.11 over multi-hop wireless networks: problems and new perspectives. Ad Hoc Networks, 2(2), 109–132. 2. IEEE Standard 802.11 (1999). Wireless LAN medium access control (MAC) and physical layer (PHY) specifications. 3. Kim, J., Crow, B., Widjaja, I., & Sakai, P. (1997). IEEE 802.11 wireless local area networks. IEEE Communications, 35(9), 116– 126. 4. Paxson, V., Allman, M., & Stevens, W. (1999). TCP congestion control. RFC 2581. 5. Fu, Z., Luo, H., Zerfos, P., Lu, S., Zhang, L., & Gerla, M. (2005). The impact of multihop wireless channel on TCP performance. IEEE Transactions on Mobile Computing, 4(2), 209–221. 6. Xylomenos, G., & Polyzos, G. C. (1999). TCP and UDP performance over a wireless LAN, IEEE INFOCOM (pp. 439–446). 7. Wu, H., Peng, Y., Long, K., Cheng, S., & Ma, J. (2002). Performance of reliable transport protocol over IEEE 802.11 wireless LAN: analysis and enhancement, IEEE INFOCOM (pp. 599– 607). 8. Balakrishnan, H., & Katz, R. H. (1998). Explicit loss notification and wireless web performance, IEEE GLOBECOM. 9. Balakrishnan, H., Padmanabhan, V. N., Seshan, S., & Katz, R. H. (1997). A comparison of mechanisms for improving TCP performance over wireless links. IEEE/ACM Transactions on Networking, 5(6), 756–769. 10. Bakre, A. V., & Badrinath, B. R. (1997). Implementation and performance evaluation of indirect TCP. IEEE Transactions on Computers, 46(3), 260–278. 11. Moon, J. C., & Lee, B. G. (2006). Rate-adaptive snoop: a TCP enhancement scheme over rate-controlled lossy links. IEEE Transactions on Networking, 14(3), 603–615. 12. Chandran, K., Raghunathan, S., Venkatesan, S., & Prakash, R. (2001). A feedback based scheme for improving TCP performance in ad-hoc wireless networks. IEEE Personal Communications, 8(1), 34–39. 13. Monks, J. P., Sinha, P., & Bharghavan, V. (2000). Limitations of TCP-ELFN for ad hoc networks, MOMUC. 14. Mathis, M., Mahdavi, J., Floyd, S., & Romanow, A. (1996). TCP selective acknowledgement options, RFC 2018. 15. Postel, J. (1981). Transmission control protocol, RFC 793. 16. Casetti, C., Gerla, M., Mascolo, S., Sansadidi, M. Y., & Wang, R. (2002). TCP Westwood: end-to-end congestion control for wired/wireless networks. Wireless Networks Journal, 8(5), 467– 479. 17. Capone, A., Fratta, L., & Martignon, F. (2004). Bandwidth estimation schemes for TCP over wireless networks. IEEE Transactions on Mobile Computing, 3(2), 129–143. 18. IEEE Standard 802.16-2004 (2004). Part 16: air interface for fixed broadband wireless access system.

R.-S. Cheng et al. 19. Durst, R. C., Miller, G. J., & Travis, E. J. (1996). TCP extensions for space communications, ACM MOBICOM (pp. 15–26). 20. Kherani, A. A., & Shorey, R. (2004). Performance improvement of TCP with delayed ACKs in IEEE 802.11 wireless LANs, IEEE WCNC (pp. 1703–1708). 21. Braden, R. (1989). Requirements for Internet hosts— communication layers, RFC 1122. 22. Network Simulator, NS-2, http://www.isi.edu/nsnam/ns. 23. Eckhardt, D. A., & Steenkiste, P. (1996). Measurement and analysis of the error characteristics of an in-building wireless network, ACM SIGCOMM (pp. 243–254). 24. Nguyen, G. T., Noble, B., Katz, R. H., & Satyanarayanan, M. (1996). A trace-based approach for modeling wireless channel behavior. In Winter simulation conference (pp. 597–604). 25. Gilbert, E. N. (1960). Capacity of a burst-noise channel. Bell System Technical Journal, 69, 1253–1265. 26. Rappaport, T. S., & Rappaport, T. (2001). Wireless communications: principles and practice (2nd ed.). New York: Prentice Hall. 27. Basagni, S., Conti, M., Giordano, S., & Stojmenovic, I. (2004). Mobile ad hoc networking. New York: Wiley–IEEE Press.

Rung-Shiang Cheng received the B.S. degree in electronic engineering from National Chin Yi University of Technology, the M.S. degree in computer science and information engineering from National Cheng Kung University, and the Ph.D. degree in electrical engineering from National Cheng Kung University in 1996, 2001, and 2008, respectively. He joined Kun Shan University as an assistant professor in the Department of Computer and Communication in August 2008. His research interests include network performance analysis, congestion control, and wireless networks. Der-Jiunn Deng received the B.S. degree in computer science from Tung-Hai University, the M.S. degree in information management from National Taiwan University of Science and Technology, and the Ph.D. degree in electrical engineering from National Taiwan University in 1997, 1999, and 2005, respectively. He joined National Changhua University of Education as an assistant professor in the Department of Computer Science and Information Engineering in August 2005. Dr. Deng is also an adjunction assistant professor of the Overseas Chinese Institute of Technology. He visited Iowa State University, USA, in 2006. His research interests include multimedia communication, quality-of-service, and wireless networks. Dr. Deng is the guest editor of the special issue on Internet resource sharing and discovery for the Journal of Internet Technology. He is now serving as a program cochair for the 2nd International Conference on Computer Science and its Applications (CSA-09). Dr. Deng is a member of the IEEE.

Yueh-Min Huang is a Distinguished Professor and Chairman of the Department of Engineering Science, National Cheng-Kung University, Taiwan, R.O.C. His research interests include multimedia communications, wireless networks, artificial intelligence, and e-Learning. He received his MS and Ph.D. degrees in Electrical Engineering from the University of Arizona in 1988 and 1991 respectively. He has coauthored 2 books and has published about 200 refereed professional research papers. Dr. Huang has received many research awards, such as the Best Paper Award of 2007 IEA/AIE Conference, Best Paper Award of the Computer Society of the Republic of China in 2003, the Awards of Acer Long-Term Prize in 1996, 1998, and 1999, Excellent Research Awards of National Microcomputer and Communication Contests in 2006. Dr. Huang has been invited to give talks or served frequently in the program committee at national and international conferences. Dr. Huang is in the editorial board of the Journal of Wireless Communications and Mobile Computing, Journal of Security and Communication Networks and International Journal of Communication Systems. Huang is a member of the IEEE as well as IEEE Communication, Computer, and Circuits and Systems Societies. Lianfen Huang received her B.S. degree in Radio Physics in 1984 and PhD in Communication Engineering in 2008 from Xiamen University. She was a visiting scholar in Tsinghua University in 1997. She is an associate professor of Communication Engineering, Xiamen University, Xiamen, Fujian, China. Her current research interests include wireless communication, wireless network and signal process.

Han-Chieh Chao is a jointly appointed Professor of the Department of Electronic Engineering and Institute of Computer Science Information Engineering, National Ilan University, I-Lan, Taiwan. He also holds a joint professorship of the Department of Electrical Engineering, National Dong Hwa University, Hualien, Taiwan. His research interests include High Speed Networks, Wireless Networks and IPv6 based Networks and Applications. He received his MS and Ph.D. degrees in Electrical Engineering from Purdue University in 1989 and 1993 respectively. Dr. Chao is also serving as an IPv6 Steering Committee member and Deputy Director of RD division of the NICI Taiwan, Cochair of the Technical Area for IPv6 Forum Taiwan. Dr. Chao is an IEEE senior member, IET and BCS Fellows.