You are on page 1of 8

2009 Fifth International Conference on Mobile Ad-hoc and Sensor Networks

Performance of TCP over Ad Hoc Networks:


Analysis and Enhancement

Mostafa Kamel 1 Gamal Abdel Fadeel 2 , Usama El Ghandour 3


Electronics and Communication Dep. Electronics and Communication Dep.
Higher Institute of Eng., EL Shorouk Academy Faculty of Engineering, Helwan University
Cairo, Egypt Cairo, Egypt.
Mostafa_kamel79@yahoo.com gam_hel@yahoo.com

Abstract—the IEEE 802.11 standard suffers from a number of In this paper, as a first step, we develop a general
serious performance problems when used alongside with TCP, methodology for analyzing the MAC-layer approaches
the most popular transport protocol, in a multi-hop wireless proposed to alleviate self-contention. In particular, we focus
on the Quick- exchange (QE), scheme which is designed
ad-hoc environment. This paper focuses on enhancement of the
with the intent of reducing the effect of inter-flow, self
performance of TCP over Ad Hoc Networks that uses the contention (e.g. between packets of the same connection
IEEE 802.11 MAC protocol. Two approaches where adopted; traveling in opposite directions), see [3]. We also analyze
the first focuses on the use of new technique in the MAC-layer IEEE802.11 with QE for a single hop as a simplified
so-called quick-exchange (QE). In the second approach, the scenario, as well as for multihop cases. Then we extend the
TCP receiver uses the delayed acknowledgement option. This analytical throughput for the multihop case by extending
paper has, focused on modeling the TCP over multi-hop IEEE over Bianchi [5] work. This is achieved by taking into
account wireless channel errors and retrials counts.
802.11 network with channel effects. The objective is to
Another approach used to improve the performance of
provide a better understanding of the effects of various TCP over Ad Hoc Networks, is the TCP with delayed ACK
parameters on TCP performance. We have present an for wireless networks. The receiver should generate minimal
analytical modeling for the system under consideration along number of ACKs required for reliable data recovery.
with numerical and simulation validations via a number of Recently, the delayed ACK strategy has been studied to
simplified examples under some simplifying assumptions. This improve TCP performance [7], [8]. However, this field
has resulted in a closed form formulas for TCP throughput is not fully exploited and many issues remain unsolved.
under various channel states. Validation results have
Some important questions include how delayed ACKs
affect TCP performance, and how to choose the optimal
confirmed the performance enhancement we have expected.
delay window in multihop wireless networks. In this paper
Keywords- IEEE 802.11(MAC), TCP, performance, we analyze the TCP throughput and delayed ACK over
Analytical modeling multihop wireless links. We will use the TCP with delayed
ACK and MAC layer with QE in order to evaluate the
I. INTRODUCTION
improvement of the performance of TCP over Ad Hoc
When TCP runs over wireless LAN (WLAN), where networks.
channel is shared among users in a multiple accesses system, This paper is organized as follows. In section II we
the "forward" TCP data and the "backward" TCP ACKs briefly describe, and analyze the DCF and QE of IEEE
compete for channel access. Data / ACK collisions over the 802.11 MAC protocols, In section III we analyze the QE of
MAC procedures may degrade the overall performance of IEEE 802.11 MAC protocols taking into account wireless
the TCP connections. On the other hand, as the IEEE 802.11 channel effect, in section IV we analyze the TCP
protocol is being standardized. A number of schemes have performance with delayed ACK in wireless networks, in
been proposed to enhance the overall performance of the section V we Analysis of TCP over IEEE 802.11 Quick
802.11 protocol. Such an, enhancement, schemes must keep Exchange under Realistic Channel conditions. Finally,
backward compatibility with the standard protocol. summery and conclusion are drawn in Section VI.
A significantly important problem in this respect is that
of predicting the interaction between the TCP protocol and II. ENHANCEMENT OF RELIABLE TRANSPORT
the 802.11 standard. In this paper, we focus, in particular, on PROTOCOL
the performance of TCP connections over Ad Hoc networks
employing the 802.11 protocol, see [1], [2], and references A. Quick-Exchange: Motivations & Overview
therein. The difficulty here is that most of these studies use Quick-Exchange (QE) [4] provides an efficient
simulation models in order to study the behavior of TCP over mechanism for ex-changing two data packets between
802.11 Ad Hoc networks. Our aim in this paper is to come adjacent nodes in the same dialogue (RTS-CTS exchange) as
up with a simple mathematical model which can be used to shown in Fig. 1. As can be seen, QE obviates RTS and CTS
predict such behavior. transmissions for DATA2. The intuition behind Quick
Exchange is to make use of the fact that the two

978-0-7695-3935-5/09 $26.00 © 2009 IEEE 136


DOI 10.1109/MSN.2009.64
DIFS Backoff SIFS SIFS SIFS DIFS
(Geometric)
Source RTS DATA 1 ACK2

ACK1

Destinati on CTS DATA 2(with ACK1)

NAV (DATA2)

NA V (DATA 1)
NA V (CTS)

Other NA V (RTS)

Zn Z n +1
Good Channel State

Figure 1. 802.11 With Quick-Exchange


Figure 2. Markov Chain model for the station
communicating nodes have agreed to communicate at a
given time. QE also eliminates the idle time due to the back- packets are transmitted simultaneously. The probability of a
off prior to the transmission of DATA2 when the IEEE transmission by the station in any arbitrary time slot is given
802.11 MAC is used. Transmission of DATA2 is free of by:
channel contention and, therefore, this reduces the incidence
ptr = 1 − (1 − τ )n (3)
of false link failures while improving end- to-end
throughput. This transmission will be successful only if one STA
has transmitted, and the others did not during the same time
B. Analytical Evaluation of 802.11 With Quick-Exchange slot. Thus, the probability of success given that the
Here, we obtain an analytical expression for the transmission is successful is,
throughput of the 802.11 with QE scheme being employed. (1 − τ )nτ (1 − τ )n −1 nτ (1 − τ )n
When QE is used with the RTS-CTS handshake mechanism. ps = = (4)
ptr 1 − (1 − τ )n
We assume, n, fixed nodes, each having a packet to transmit
at all times (saturation conditions). We, also, assume perfect The time spent for a successful transmission of the DATA
channel conditions and no hidden terminals [5]. This packet by the STA is calculated as:
assumption will be removed later. Let us begin, by defining T s = o rts + {H + E d + δ + SIFS + H + E a + δ +
the system throughput as follow, SIFS + ACK + δ + DIFS }
E [Payload information transmitted in a slot time] Where
S= (1)
E[Length of slot time] o rts = RTS + SIFS + CTS + SIFS + 2δ
As illustrated in Fig. 2, following Bianchi [5], we In case of a collision, when RTS/CTS scheme is used, the
assume that each station transmits in a randomly chosen collision time is given by
time slot with probability τ and collisions occur with a T c = RTS + δ + DIFS
constant probability, p irrespective of the number of Where S is the system throughput as derived by Bianchi [5]:
previous collisions before any successful transmission. We ptr × ps × [E d + E a ]
consider the time slot to be of length σ . S perfect channel = (5)
(1 − ptr )σ + ptr psT s + (1 − p s ) ptrT c
Let E d (E a ) , H and δ denote respectively, the average
Fig. 3 shows the numerical results for the throughput
packet's payload size, the packet header's size (denoted along with the simulation results obtained by the simulation
as PHY hdr + MAC hdr ). And the propagation delay. model we have developed for this purpose. As can be seen,
E d (E a ) And H is measured in unit time. For simplicity we QE improves the throughput performance of the Ad Hoc
define, CW = CWmin and m as the maximum backoff network.
slots so that, CWmax = 2m . CWmin .The probability of D. Delay Analysis
transmission, τ in a given time slot is given by [5], The mean delay, D, can be obtained by computing the
2(1 − 2 p )(1 − p ) time interval between two successful transmissions at a
τ= (2) node, it can be calculated directly from the throughput as
(1 − 2 p )(cw + 1) + p (cw (1 − (2 p ) m )
follows [10],
In what follows, we identify various events that can
n .E
occur in an arbitrary time slot under Quick Exchange D= (6)
operation and calculate their occurrence probabilities. S
Fig. 4 shows the numerical and simulation results for
C. When station contends for the channel the Delay as function of number of stations. As expected,
In this case, the station contends for the channel with all the QE results in a better delay performance than that of the
other nodes. In this scenario, QE works similar to DCF and DCF.
collisions can occur when two or more RTS

137
Figure 5. Model of wireless channel

• Throughput analysis:
Figure 3. Throughput as a function of number of stations (n)
While the basic analysis methodology here is the same
as that used by Bianchi [5]. However, we take into
consideration, the (Good/Bad) channel state.
Each node listens to the same channel in the collocated
case. Fig. 5, models different channel states with λ g and λb
designating the transition probabilities between the good
(bade) states respectively. Here, the wireless channel state at
any given time is undergoing one of the following eight
events: the key idea is to compute the probabilities pi for
each of the (eight) events in terms of τ , λ g and λb , and

Figure 4. Delay as a function of number of stations (n)


then compute the overall throughput,
1) Idle slot : If none of the nodes transmits a RTS frame.

TABLE I. PARAMETERS USED TO OBTAIN SIMULATION


p1 = (1 − τ ) n (7)
AND NUMERICAL RESULTS d1 = σ
Parameter
P t
Value
Where: pi Probability of each events, d i duration of
SIFS
DIFS
28µs
128µs each events, σ time slot
channel bit rate 1Mbit / sec
2) RTS – RTS Collision: If more than one station
Mac header 272 bits
transmits a RTS in the same slot. So,
PHY header 128 bit
TCP DATA 8192 bits
P2 = 1 − (1 − τ ) n − nτ (1 − τ ) n−1 (8)

TCP ACK 1024bits


d 2 = RTS + δ + EIFS
propagation delay 1µs 3) RTS Corrupted: The probability for RTS corruption
ACK 112bit+PHY given that there is no RTS collision is calculated as follows:
h d
RTS 160 bit+PHY λb − λ g ( RTS +δ )
CTS
h d
112bit+PHY P3 = nτ (1 − τ ) n−1 (1 − e ) (9)
h d λg + λb
CW 16
d 3 = RTS + δ + EIFS
Backoff stage 6
Time slot 5 µs 4) CTS Corrupted: Given that some RTS gets through
successfully, is given by,
III. MAC/QE PERFORMANCE UNDER REALISTIC λb − λg ( RTS +δ )
CHANNEL CONDITIONS P4 = nτ (1 − τ ) n −1 e .
λg + λb (10)
This section presents analytical model for the T A (SIFS ) T A (SIFS ) − λg (CTS +δ )
performance of IEEE 802.11 wireless Ad Hoc network with [(1 − B e B)+B e B (1 − e )].
QE scheme, under realistic channel conditions [6].
d 4 = RTS + δ + SIFS + CTS + δ + EIFS
A. Performance analysis of a single hop 802.11 wireless
In this section, we model a wireless Ad Hoc network
where the nodes are all collocated, and therefore, frames
need only traverse one hop before reaching their destination.

138
5) Data-1, Corruption: ptr ′ × pS ′ × [E d + E a ]
n −1 λb − λg (R T S +δ ) S Realistic Channel = (15)
p 5 = n τ (1 − τ ) e . (1 − ptr ′ )σ + ptr ′ pS ′T s + (1 − pS ′ ) ptr ′Tc
λ g + λb
Where
{1 − [(1 -B T e A (S IF S ) B ) + B T e A (S IF S ) B 2(1 − 2 p ′)(1 − p ′)
τ′ =
(1 - e
- λ g (C T S + δ )
)]} .[(1 -B T
e A (S IF S )
B) (1 − 2 p ′)(cw + 1) + p ′(cw (1 − (2 p ′) m )

+B T
e A (S IF S )
B (1 - e
- λ g (d a ta + δ )
) ] (11) ptr ′ = 1 − (1 − τ ′)n
pS ′ = p8
d 5 = RTS + δ + SIFS + CTS + δ + SIFS +
In Fig. 6, we can see that the use of Realistic Channel
Data1 + δ + EIFS
led to the decline in the Throughput, but we have noticed
6) Data-2, Corruption: that QE is still the best of the DCF. In Fig. 7 we note the
λb − λ g ( RT S + δ )
effect on the Throughput at different state of channel. In
p 6 = nτ (1 − τ ) n −1 e . Fig. 8 We note that the delay also affected if the Realistic
λ g + λb Channel condition and it increased on the state of the perfect
{1 − [1 − [(1-B T e A(SIFS) B) + B T e A(SIFS) B channel in Fig. 4.
- λg (CTS + δ )
(1 - e )]].[(1-BT e A(SIFS) B) + BT e A(SIFS) B
- λg (data 1+ δ )
(1 - e )]}.[(1-BT e A(SIFS) B) + B T e A(SIFS) B
- λ (data 1+ δ )
(1 - e g )] (12)
d 6 = RTS + δ + SIFS + CTS + δ + SIFS + Data1 +
δ + SIFS + Data 2 + δ + EIFS
7) ACK Corruption:
λb − λ ( RTS +δ )
p 7 = nτ (1 − τ ) n −1 e g .
λg + λb
Figure 6. Throughput as a function of number of stations (n) with
{1 − [1 − [1 − [(1-BT e A(SIFS) B) + BT e A(SIFS) B Realistic Channel
- λg (CTS+δ )
(1 - e )]].[(1-BT e A(SIFS) B) + BT e A(SIFS) B
- λg (data1+δ )
(1 - e )]].[(1-BT e A(SIFS) B) + BT eA(SIFS) B
- λg (data2 +δ )
(1 - e )]}.[(1-BT e A(SIFS) B) + BT e A(SIFS) B
- λg (ACK +δ )
(1 - e )]. (13)
d 7 = RTS + δ + SIFS + CTS + δ + SIFS + Data1 +
δ + SIFS + Data 2 + δ + SIFS + ACK + δ + EIFS
8) Successful: Finally, the probability of successful
transmission is Calculated from the eq. 7 to eq. 13 as Figure 7. Throughput as a function of number of stations (n) for QE in
follows: perfect channel and Realistic Channel
p8 = 1 − (P1 + P2 + P3 + P4 + P5 + P6 + P7 )
(14)
d 8 = R T S + δ + S IFS + CT S + δ + S IFS + Data1 +
δ + S IFS + Data 2 + δ + S IFS + A CK + δ + DIFS
Where the probability p ′ that the RTS transmission is
unsuccessful given that the node transmits a RTS can be
computed in terms of τ ′ in exactly the same manner as we
did above,
p ′ = p 2 + p 3 + p 4 + p5 + p 6 + p 7
Hence, the system throughput, S, under realistic Figure 8. Delay as a function of number of stations (n) with channel
Channel conditions is: effect

139
IV. PERFORMANCE OF THE TCP OVER WIRELESS
AD HOC NETWORKS
In the following, we examine the performance of TCP
over Ad Hoc networks with QE We now propose a general
method for modeling TCP over wireless Ad Hoc networks.
This analysis gives a closed form expression for TCP
throughput under the simplifying assumptions such as [2].
A. Description of the Setup
Fig. 9 depicts an 802.11 wireless LAN. There are 3
nodes, Node 2 is in the transmission range of both node 1 Figure 10. Showing the Markov chain {P1 , S , P3 } For W=2 case.
and 3. Nodes 1 and 3 are not in each other's transmission
range. For the time being, assume that the transmission • R n Denotes a random variable which is d if an
range and the sensing range are same. The nodes are ACK packet has been transmitted from node 2 to
assumed to be static which implies that the routing is also node 1 in interval ( Z n , Z n +1 ) and R n = 0
static. The MAC layer implements the four way handshake
otherwise.
for data transmission. Further, for ease of presentation, we
assume that TCP's window size is fixed at W. The channel C. TCP(Delay ACK) Throughput over Quick Exchange for
is also assumed to be in a good state so that any W=d=2:
transmission is not corrupted by channel errors; this We apply the case where the receiver (node 3) employs
assumption will be removed later. delayed ACK option of TCP with the delayed ACK
Following [2], where it is focusing on calculating the parameterW = d , and assume that W=d=2 (the case of
throughput for the TCP over IEEE802.11b for the DCF. But general W requires some minor approximations and is
we will extend the analysis to the QE case.
studied next). Fig. 10 depicts the Markov chain {P1 , S , P3 }
B. Performance Enhancement of TCP with Delayed ACKs describing the system states. In the Fig. 10 the transitions
in IEEE802.11b Wireless Ad Hoc Networks indicated with dashed lines ( B → C and B → D ) are with
This section studies the TCP performance with delayed probabilities 0.5 and those shown with solid lines are with
ACK in wireless networks which use IEEE 802.11 MAC probabilities 1.
protocol as the underlying medium access control. Recently, A simple computation shows that the stationary
the delayed ACK strategy has been studied to improve TCP distribution of the Markov chain is:
performance [7], [8]. We assume that d =W which is ⎧1
expected to give best throughput performance for a given ⎪⎪ ,.......X = C , D
window size W. π (X ) = ⎨10 (17)
Now, the throughput achieved by the TCP session is ⎪ 1 ,.........Otherwise .
E [R ] ⎪⎩ 5
(16) Note that R n = 2 only for the transitions F → A and
E [L ]
1 n R = 0 otherwise. Also, the mean time between two
Where E[R] = lim n →∞ ∑ R i successive successful transmissions is (assuming no
n i =1
collisions):
1 n
And E [L ] = lim n →∞ ∑ Z i +1 − Z i E[Z n+1 − Z n ] = DIFS + Tb + RTS + 3SIFS +
n i =1
• Z n Denotes the time instant at which n th successful CTS + TTD + ACK
transmission gets over in the network, i.e., at instant E [Z n +1 n
− Z ] = T o +T b + TTD (18)
Z n the n th MAC layer acknowledgement is To = DIFS + RTS + 3SIFS + CTS
received by one of the three nodes.
Let T x = 3SIFS + ACK
TCP Source TC P Sink
So; E [ Z 'n +1 − Z 'n ] = T X + T TA
Where T b is the expected backoff time for the
1 2 3
transition. Note that, under assumption of geometrically
Figure 9. Showing a line network with 3 nodes. distributed backoff times and no exponential bakeoffs,
δ
Tb = for transitions from a state where a single node has
p

140
δ
both the packets, and T b = for transitions from a state
2p
where two distinct nodes have packets to be transmitted,
1 CW
where = . Let TTD = H + E d andTTA = H + E a ,
P 2
denote respectively, the TCP DATA and TCP ACK.
It follows, after some simple algebra, that TCP
throughput with delayed ACK over Quick Exchange is
given by
E [R ] 2 Showing the Markov chain {P1, S , P3} for general W
S Q_E = = (19) Figure 11.
E [L ] 9δ
5T o + 4TTD + 2TTA +
2p true because the backoff period is negligible as compared to
Based on similar analysis in [2], one can show that the the rest of the total TCP DATA transmission time. It follows
throughput of TCP with delayed ACK over DCF with that,
W=2and d=2 will be, E [L ]  2π (W ,1′)E [TA ] + (1 − 2π (W ,1′))E [T D ]. (21)
2 For the case where no collision occurs, with Quick
S DCF = (20)
11δ Exchange:
6T o + 4TTD + 2TTA +
2p E [T D ] = DIFS + E [T B ] + RTS + SIFS + CTS +
(22)
By comparing (19) and (20), we can see that there is a SIFS + TTD + SIFS + ACK
significant improvement when the receiver employs the Similarly,
delayed acknowledgement option with Quick Exchange, and E [T A ] = TTA + SIFS (23)
reduced control overheads as well as reduced average time
that a node spends backing off before transmission. Thus the expected throughput for a window size of W is,
(noting that π (W ,W ) = π (W ,1' ) )
D. Analysis for General Window size
W
It has been observed [9], that in a chain network, a S (W ) =
standard TCP connection (i.e., no delayed 2(E [T A ] +W E [T D ])
acknowledgements) with fixed window size of W packets From Table 1, above, we can calculate E [T A ] and E [T D ] ,
attains a maximum throughput if W is one fifth of the and hence,
number of hops in the chain. While [7] suggested that by W
using delayed acknowledgements, one can improve the S (W ) Perfect channel = (24)
performance seen by a TCP connection over an Ad Hoc (N − 1)(E (T A ) +W E (T D ))
network using the IEEE 802.11 MAC layer protocol. The analytical throughput, S (W) can be obtained from
In this section we assume that the delayed ACK Equation 16 with the number of nodes (3). W is the (fixed)
parameter is d=W. Fig. 11 depicts the evolution of the TCP window size. Fig. 12 shows the throughput achieved
Markov chain {P1 , S , P3 } for the general window size case. by TCP session with DCF and QE. From analysis and
simulations, which by using the delayed acknowledgement
As can be seen, one TCP ACK is generated after every
option with QE, the TCP performance can be improved
2W+2 transitions (2W for transition from state (0, 0) to (W,
significantly by increasing the TCP's window size. This gain
W) and the other 2 transitions for (W, W) to (0, 0)), it
in performance was obtained by reducing the feedback
follows that the stationary distribution of the Markov chain
traffic which, in the given network, competes with the TCP
1
is π (W ,1' ) = . Thus the expected inter- DATA packets for transmission in the forward direction
2W + 2
acknowledgement time is,
i ,W −1)
E [L ] = 2π (W ,1′)E [T A ] + ∑Wi=0 ∑ min(
j =0 π (i , j )E i , j [T D ]
Where E i , j [T D ] denotes the mean transfer time required
to transfer a TCP DATA packet in state (i , j ) and E [T A ] is
the expected time for transmission of a TCP ACK packet.
• An approximation:
In the following, we assume that E i , j [T D ] is state
independent of (i,j) and is denoted by E [T D ] ; this can be
Figure 12. Show the TCP with delay ACK (over DCF and QE)
throughput for three (N=3) nodes.

141
V. ANALYSIS OF TCP OVER IEEE 802.11 QE UNDER
REALISTIC CHANNEL CONDITIONS
We consider the scenario depicted in Fig. 8, and
detailed in Section V. We assume that the channel state is
time varying (the exact dynamics of the channel states is
given in Fig. 5). This modification in the system requires
some changes in the variables we have defined previously,
as follows.
• The Markov Chain
At instant Z n , all the nodes that have packets to
transmit. Start decreasing their residual backoff timers. Thus
this would seem as if all nodes will start their first attempt Figure 13. the state transition diagram of the Markov
of transmission. Thus we can use the standard Markov chain {P1( n ) , S ( n ) , P3( n ) }
approach to find the TCP throughput.
Where
A. The case of λ g + λb = 1
E [ Z ] = DIFS + E [T B ] + (1 − λb RTS )RTS +
Let us assume that the TCP receiver (node 3) employs
λb RTS (1 − λb CTS )(RTS + SIFS + CTS ) +
the delayed ACK option of TCP with d=2. Recall that the
TCP window size is also fixed at W=2. λb RTS +CTS (1 − λbT D )(RTS + SIFS + CTS
• For Quick Exchange:
+SIFS +T D )π (A ) + λb RTS +CTS +T D (1 − λb ACK )
Fig. 1 shows the requirement on the channel state for (RTS + SIFS + CTS + SIFS + T D + SIFS + ACK ).
the transmission between instants Z n and ( Z n , Z n +1 ) to be And
successful. The requirement is that the channel state must be E [Z `] = DIFS + E [TB ] + (1− λb RTS )RTS +
good in the intervals marked as good channel. Also, the
length of the (Geometric) backoff timer value depends on λb RTS (1− λbCTS )(RTS + SIFS +CTS ) +
the number of nodes attempting a transmission. Thus, if the λb RTS +CTS (1− λbTD )(RTS + SIFS +CTS + SIFS +
transmission is of TCP DATA packet, the probability of
successful transmission is ( λb the probability of good TD )π (A ) + λb RTS +CTS +TD (1− λbTTA ).(RTS + SIFS +
channel state): CTS + SIFS +T D + SIFS +TTA )π (F ) + λb RTS +CTS +TD +TTA
pd = λb RTS +CTS +TTD + ACK (25) (1− λb ACK ).(RTS + SIFS +CTS + SIFS +T D + SIFS +
And if it is TCP ACK packet, the probability is: TTA + SIFS + ACK )
pa = λb RTS +CTS +TTA + ACK (26)
E [Z ``] = DIFS + E [T B ] + (1 − λb RTS )RTS + λb RTS
Fig.13 depicts the evolution of the Markov
(1 − λb CTS )(RTS + SIFS + CTS ) + λb RTS +CTS
chain {P1( n ) , S ( n ) , P3( n ) } , at the instants when the nodes
(1 − λbTTA ).(RTS + SIFS + CTS + SIFS +TTA )π (F )
having packet to be transmitted start decrementing their
backoff timers (after the senders have sensed an idle channel +λb RTS +CTS +TTA (1 − λb ACK ).(RTS + SIFS + CTS +
for DIFS period). SIFS +TTA + SIFS + ACK ).
• TCP Throughput over QE under Realistic Channel: Computing: E [T B ] The expected backoff time value is
Note that it is possible that the transmission in a given 1 1 1 π (B )
channel state may not be successful. In that case the length E [T B ] = π (B ) + (1 − π (B )) = (1 − )
2p p p 2
of the period until Ack is received depends on the location
The stationary distribution of the Markov chain of Fig. 10 is
of the first occurrence of a bad state along the intervals
given by,
required to get in a good state with a successful
transmission. The channel is in good state for all the packets 1 − pa
π (A ) = π (B ) = 2π (C ) = 2π (D ) = π (E ) =
transmitted after and before successful transmission and the 5 − pd − 4 pa
channel is in bad state at least once during successful 1 − pd
transmission. And π (F ) =
5 − pd − 4 pa
The expected period length until ACK is received,
Rn is equal to the number of TCP DATA packets that
E [L ] = 3E [ Z ] + E [ Z `] + E [ Z ``] (27)
a TCP ACK received at node 1 acknowledges

142
and applied our findings to illustrate the effect of ACK
thinning (due to delayed acknowledgement ). The results
have confirmed a significant improvement in TCP
performance as the delayed acknowledgement
parameter, d, increases to the TCP window size This
paper have, also, proposed a new and simple analytical
model based on Markov chain to compute the
throughput performance of IEEE 802.11 with Quick
exchange and TCP with delayed acknowledgement.
Comparisons have shown that this model is accurate in
Figure 14. Show the throughput of TCP with delay ACK (over DCF and predicting the 802.11 system's throughput. Taking into
QE) with Realistic Channel for (N=3). account, the (good/bad) wireless channel states and the
analytical modeling we have presented along with the
corresponding validations is considered to be a
significant contribution for the work we have carried out
in this paper
The final conclusion we should make is that, Quick
exchange and delayed acknowledgement can be used
together to enhance the performance of TCP over Ad
Hoc networks. We believe that the paper is an important
step towards understanding the interaction between the
TCP protocol and Ad Hoc wireless networks wherein
nodes are employing the IEEE802.11 MAC.
Figure 15. Show the throughput of TCP with delay ACK (QE) in perfect References
channel and Realistic Channel for (N=3).
[1] H. Wu, Y. Peng, K. Long, S. Cheng and J. Ma, “Performance of
i.e., Rn = 2 for F → A transition and is 0 otherwise. Reliable Transport Protocol over IEEE 802.11 Wireless LAN:
Analysis and Enhancement,” IEEE INFOCOM'2002
The expected TCP ACK received is: [2] Arzad A. Kherani,and Rajeev Shorey, " Simple Markovian Models
E [R ] = 2π (F )λb RTS +CTS +TTA + ACK (28) for Performance Evaluation of TCP with in IEEE 802.11b
Wireless LANs ",in Proceedings of IEEE Wireless
The ratio of E [L ] and E [R ] is easily obtained from Communications and Networking Conference, pp. 237–242, 2004
Equations (27) and (28), thus giving the throughput [3] D. Berger et al., “TCP-Friendly Medium Access Control for Ad-
Hoc Wireless Networks: Alleviating Self-Contention,” MASS ’04,
achieved by TCP. Fig. 14 shows the throughput achieved by
Oct 24-27, 2004, Florida
TCP session with DCF and QE with Realistic Channel [4] D. Berger, Z. Ye, P. Sinha, S. Krishnamurthy, M. Faloutsos, and
condition. It is seen that using the delayed ack option with S. K. Tripathi. Alleviating MAC Layer Self-Contention in Multi-
QE results in significant improvement in the throughput hop Wireless Networks. Technical Report, Dept of CS, UC
Riverside, 2003.
performance but it decrease than perfect channel. Fig. 15
[5] G.Bianchi. Performance Analysis of the IEEE 802.11 Distributed
shows different cases of the channel effect on the TCP Coordination Function. IEEE Journal on Selected Area in Comm.
(Delayed ACK) Throughput over Q E. V18, N3, March 2000
[6] Nitin Gupta and P.R. Kumar '' A performance Analysis of the
VI. SUMMARY AND CONCLUSIONS 802.11 Wireless LAN Medium Access Control ", Vol. 3, No. 4,
pp. 279-304, September 2004
The performance of a reliable transport protocol, [7] E. Altman and T. Jimenez, “Novel delayed ack techniques for
such as TCP, over Ad Hoc network needs careful improving tcp performance in multihop wireless networks,”
studies. TCP needs the transport layer acknowledgement Personal Wireless Communications, September 2003.
[8] R. de Oliveira and T. Braun, “A dynamic adaptive
(TCP ACK) on the backward direction. In the scenario acknowledgment strategy for tcp over multihop wireless
of TCP over Ad Hoc network where a shared channel is networks,” Infocom, 2005.
used in a multiple access, the forward TCP data and the [9] K. Chen, Y. Xue, and K. Nahrstedt `On Setting TCP's Congestion
Window Limit in Mobile Ad Hoc Networks'. ICC’03.
backward TCP ACK will compete for the channel, [10] Bo Li, Roberto Battiti, "Performance analysis of an enhanced
which may cause collisions and degrades the overall IEEE 802.11 distributed coordination function supporting service
performance. Based on these observations, this paper differentiation", QoFIS, in: International Workshop on Quality of
Future internet Services, 2003, Sweden, LNCS 2811, pp. 152-161.
proposed an analytical derivation for the quick-exchange
scheme which facilitates efficient exchange of packets
between communicating neighbors, reducing inter-flow self
contention. We have, also, proposed a general
methodology for modeling TCP over 802.11 networks

143

You might also like