Content uploaded by Khaled Elleithy
Author content
All content in this area was uploaded by Khaled Elleithy
Content may be subject to copyright.
Simulation based Study of TCP Variants in Hybrid Network
Wafa Elmannai, Abdul Razaque and Khaled Elleithy
welmanna@bridgeport.edu arazaque@bridgeport.edu elleithy@bridgeport.edu
Department of Computer science and Engineering
University of Bridgeport, Bridgeport, CT 06604
Abstract:
Transmission control protocol (TCP) was originally
designed for fixed networks to provide the reliability of
the data delivery. The improvement of TCP performance
was also achieved with different types of networks with
introduction of new TCP variants.
However, there are still many factors that affect
performance of TCP. Mobility is one of the major affects
on TCP performance in wireless networks and MANET
(Mobile Ad Hoc Network).
To determine the best TCP variant from mobility point of
view, we simulate some TCP variants in real life scenario.
This paper addresses the performance of TCP variants
such as TCP-Tahoe, TCP-Reno, TCP-New Reno, TCP-
Vegas, TCP-SACK and TCP-Westwood from mobility
point of view.
The scenarios presented in this paper are supported by
Zone routing Protocol (ZRP) with integration of random
waypoint mobility model in MANET area. The scenario
shows the speed of walking person to a vehicle and suited
particularly for mountainous and deserted areas. On the
basis of simulation, we analyze Round trip time (RTT)
fairness, End-to-End delay, control overhead, number of
broken links during the delivery of data. Finally analyzed
parameters help to find out the best TCP variant.
I. INTRODUCTION
The fact that TCP is originally started with the wireless
network, even its well performance in wired network. [16]
As well as it is cleared that the deployment of wireless
networks in the past few years have motivated lot of
people to study and make efforts for improving the
performance of TCP in wireless networks and all this
work confirmed that TCP in its present structure is not
appropriate for MANET. Since the TCP has over head of
packet loss which is may cause the buffer congestion
because the MANETs losses is due to error or the
frequently of the mobility .So, the TCP can’t be efficient
probably [15].
Furthermore, the main reason of the weak performance of
the standard TCP in the wireless network is the inability of
discovers the packet loss. So, we can realize that both
characteristics have same reason of packet loss problem
which is basic on the network congestion. [16]
However, some new protocols have been proposed and
implemented. We will evaluate some of these protocols,
and we will demonstrate how they increase the
performance on wireless networks. This study will be
based on Hybrid network particularly MANET.
By going back to the MANET which is also has increased
due to the spreading of inexpensive portable and
computing devices. The MANET network is special
network due to the ability of its nodes to communicate
with each other through packed forwardly by the
intermediate nodes. So, it can be set up in any remote
areas without infrastructure support. The nodes, which are
part of MANET and they require data and information
from database but database is available in the wired
network, therefore MANET can be integrated with wired
network to obtain the[14] required data and information.
Some applications run over the database and these
applications are supported by Transmission Control
protocols. This paper aims to exhibit the flaws for TCP
compare to Hybrid network especially MANET.
Exhibition will be done with ns2. This simulator will
provide the outcomes for different protocols’ throughput,
broken links overhead etc…that we will analyze. TCP uses
some congestion control parameters, which include
congestion window, recovery mechanism, retry limit,
maximum packet size and back up mechanism for IEEE
802.11 retransmission [2, 5]. To minimize the congestion
problems, as different TCP Variants have been introduced
& simulated on various schemes in order to identify the
performance for each TCP Variants and analyzing which
variant has considerable performance due to mobility. To
determine the performance for each TCP Variants, as new
architecture and approaches are required to find out
complete behavior of the variants. This motivation results
to introduce such network to analyze the effectiveness for
each TCP Variants. The random waypoint mobility model
is incorporated to control the moments of nodes. In order
to analyze the impact of mobility while simulating the
Hybrid network, it is essential that underlying mobility
model attain realistic scenario or at least important feature.
To this conclusion, we deem that this paper makes
reasonable contribution. The rest of the paper is organized
as follows. In Section II, We describe problem statement.
In Section III, we present related work. In Section IV, we
present Design of Hybrid network Scenario. In Section V,
we define Setup of Initial connection and Hand off
process. In Section VI, We give Overview of mobility
model and simulation setup. In Section VII, We present
simulation Results, In Section VIII; we talk about
discussion of Results. Finally Section IX concludes the
work and future directions.
II. RELATED WORK
Ramarathinams et al. [18] evaluated the performance of
TCP Reno, New Reno, SACK and Tahoe with respect to
Goodput under three routing protocols over static multi-
hop network and claimed that Reno got better results but
scenario is not fully explain. Our work is completely
different from their work. We introduce ZRP protocol in
Hybrid network with inclusion of additional TCP Vegas
and Westwood. Our work almost discusses all the issues
of Manet due to TCP Variants and routing protocol. Abdul
Razaque et al. [7] compared TCP Variants in APN Hybrid
network by using DSR routing protocol. We previously
focused on Throughput, Packet delivery ratio and End-to-
End delay but here point out the performance of some
existing TCP Variants from different angles and
incorporated Hybrid routing protocol “ZRP” in our
architecture. We focus on control overhead, In-order
delivery of data, broken links, RTT from different
perspective. The finding of this paper gives complete
knowledge about the behavior of each TCP Variants in
Hybrid environment. We narrowly analyze all issues of
Manet due to mobility and showed their affect on the
performance of each TCP Variants.A.O. Oluwatope et al.
[5] Used the realistic scenario of Hybrid network and
simulated the TCP Reno, TCP SACK and TCP
Westwood. They claimed that TCP Westwood was better
performer in their static scenarios whereas our work was
completely depends on mobility and speed with Random
way point model. We have thoroughly studied the
behavior of TCP Variants.
III. PROBLEM STATEMENT
The scope of this paper is to analyze some existing TCP
Variants over Hybrid network. The focus of study is
particularly around the performance metrics such as
throughput, RTT fairness, End-to-End delay, broken links
and control overhead due to mobility. The major
contribution of this research is to identify the loss of
Goodput on different mobility ratios and to design
mobility based Hybrid network with Random Waypoint
Mobility model, where TCP Variants will be simulated
and analyzed from mobility point of view.
IV. DESIGN OF HYBRID NETWORK
SCENARIO
We have designed random waypoint mobility aware
scenario in Hybrid network by combining the features of
wired network with wireless and MANET in order to
make reasonable communication even in remote areas.
The nodes, which make the possible communication
between different segments of network are called gateway
(Anchor Point Node).
The APN can play a role as coordinator in the network.
Three segments of networks are jointly connected to make
the Hybrid network. The APNs are located on different
positions. The gateway(APN) of MANET has information
about the nodes, and these nodes are assigned the IPs
locally through Dynamic Host Configuration Protocol
Server (DHCP). An APN that is part of MANET is said to
be MANET Anchor Point Node (MAPN) similarly, the
node that is located at the area where wireless range
becomes weak is called Infrastructure Based Anchor Point
Node (IBAPN). Both APNs can play a role as coordinators
and formulate possible communication for rest of nodes in
fixed and MANET segment of network. The wired and
wireless segments of network cover the urban and suburbs
areas of urban environment and MANET portion of
network covers remote areas. We use random waypoint
(RW) mobility model at different mobility ratios and
speeds in this network. The MANET network is routed
with ZRP. We have created twenty traffic flows to analyze
the performance of TCP Variants. This Hybrid network
could be suited for urban and remote areas given in
figure1.
Figure 1: Design of Hybrid Network Scenario [3]
V. SETUP OF INITIAL CONNECTION AND HAND
OFF PROCESS
The section presents outline of initial connection setup and
handoff process for Manet Mobile Node (MMN). Figure 2
shows timing diagram and describe the signals involved in
it. Initial connection setup and MMN hand off process can
be defined in the following steps. Initially the nodes,
which are the part of Manet, intending to communicate
with corresponding node (CN). They should establish
initial connection setup and send the message through
Current MANET Anchor Point Node (CMAPN) “Request
for connection setup with CN”.
1. When CMAPN obtains the Request for connection
setup from MMN and forwards the message “coordination
request for connection setup” to Infrastructure Based
Anchor Point Node (IBAPN). In response CMAPN also
sends back message “Reply for connection setup with CN”
to (MMN). When MMN obtains the message from
CMAPN then it will be waiting till initial connection is
established. 2. IBAPN forwards the message “forwarding
coordination request for connection setup” to respective
HA/FA within wired area. HA/FA informs the IBAPN
with message “Accept coordination request” to CMAPN.
3. HA/FA forwards message with “forwarding initial
connection setup” to (CN) and sends back response to
IBAPN “Accept forwarding coordination request”. When
CN receives the message then inform the HA/FA “Accept
initial connection setup” with (MMN). With establishment
of initial connection setup between CN and MMN then
data exchange process is started.
4. When MMN changes the location and moves to other
MANET then it sends the request for handoff to new
MANET Anchor point node (NMAPN) with message
“request for joining”.
5. NMAPN sends the message “location change
forwarding message” (LCFM) to IBAPN for informing the
handoff process and similar message is forwarded to
HA/FA and finally to CN for location update.
6. NMAPN forwards the LCFM to IBAPN and also
“update” to CMAPN. In response, CMAPN sends
acknowledgement (ACK) to NMAPN for location update.
7. When CN gets the message LCFM then it sets the
connection again with MMN and message is forwarded
with “new connection setup in change of location”.
8. With the establishment of new connection, the data
exchange process is initiated.
Figure 2: Initial connection setup and Hand off
process [1].
VI. OVERVIEW OF MOBILITY MODEL &
SIMULATION SETUP
The purpose of this paper is to investigate the results of
TCP Variants under mobility Based Hybrid Network. In
this paper, we have critically evaluated the Performance of
TCP Variants with respect to different mobility ratios and
analyzed the behavior of each TCP Variants. The
parameters of interest include throughput, RTT fairness,
In-order delivery of data, effect of mobility on Goodput,
End-to-End delay and control overhead.
VI. (A) OVERVIEW OF RANDOM WAYPOINT
MOBILITY MODEL
The literature survey gives the detail of mobility models in
[6] but these mobility models mostly described
theoretically and major variation in mobility patterns are
found in real scenarios. The new mobility models were
also introduced in [9], such as free Mobility models (FM),
Manhattan Mobility model ((MM) and Reference Point
Group Mobility model (RPGM). The social network
Mobility model (SNM) is discussed [7].The research
community mostly uses the Random Waypoint (RW)
Mobility model in Mobile Adhoc networks. RW Mobility
model is also incorporated in ns2 and detail of this model
is given in following Para: The nodes move to random
destination with given velocity by using normal or uniform
distribution [Velocity minimum, Velocity maximum]
when nodes reach the destination, they stop for the time
given by the “pause” time. The pause time can be constant
value or uniform distribution [0, time pause maximum].
After completion of pause time, mobile nodes decide the
destination and direction randomly and this process
continues till the simulation time ends.
VI. (B) SIMULATION SETUP
Ns2.28 on Red Hat 8 is used for simulation. The random
Waypoint mobile scenario is generated. The simulator
gives a proper model for signal propagation and
transmission range is 250 meter [8]. The sensing and
interference range is 550 meter. TCP New Reno, Reno,
Tahoe, SACK, Westwood and Vegas are simulated and
investigated on the same network so as to ensure fairness
and behavior of the TCP Variants. The length of packet is
1040 bytes including 40 bytes are overhead. In this
simulation, 40 mobile nodes both in wireless and MANET
segment of network are placed. As we check the mobility
of MANET-nodes, which move within rectangular field of
600 *1200 meters. RW generates mobile scenario and start
location of nodes. Constant values for pause time have
been set, which are 10 seconds after each 50 seconds.
Total simulation time is 300 seconds. The minimum speed
of the node (Vmin) is 0 m/sec and maximum speed
(Vmax) are 10 m/sec respectively. The moving speed of
node is randomly obtained through uniform division
[Vmin, Vmax]. We run simulations, which cover
combination of the pause time and moving speed of nodes.
The percentage of mobility means how many mobile
nodes move and resulting how many links break in the
MANET. Hence 50% mobility shows 20 nodes move out
of 40 nodes. Zone Routing Protocol (ZRP
is locally
proactive and globally reactive , which gives better
performance for routing in multi-hop Mobile Adhoc
Network (MANET) and produces minimum routing
overhead. It has also capability to deliver approximately
all originated data packets, even with perpetual, rapid
movement of all nodes in the network. The major cause
for better performance is that ZRP functions completely on
demand with no periodic motion of any type mandatory at
any stage in the network.
VII. SIMULATION RESULT
In this subsection, we discuss the results of simulated
scenario.
A. (THROUGHPUT FOR TCP VARIANTS AT
DIFFERENT MOBILITY RATIOS)
We have simulated Hybrid network Scenario with network
simulator-2 and analyzed the performance of TCP
Variants (Reno, Vegas, New Reno, Tahoe, Westwood and
SACK). We have collected acknowledged packets for
each TCP Variants and analyzed the throughput
performance. The figure 3 shows the throughput
performance for each TCP Variants at the maximum speed
of 5 m/sec with Random Waypoint Mobility model. The
performance gradually decreases to each TCP Variants
from 5% mobility to 50% mobility. The reasons for
decreasing rate of delivered packet is mobility since 2
nodes move in the network that less links breaks and takes
less time to recover whereas 20 mobility nodes cause more
time to recover from broken links. If MANET segment of
network is part of Hybrid network, the topology of
MANET remains mostly dynamic and major affecting
factors are radio channel fading and mobility of nodes
[15].The mobility also degrades the performance of TCP
Variants because mobility causes the change of routing
information in network, which causes long RTT and
repeated timeouts resulting takes long time in retaining.
Due to mobility, the receiver gets out of order segments
resulting in the receiver generates Acknowledgements
(Ack) only for highest in-order packets. This causes the
duplicate Acks and fast retransmission algorithm starts
and congestion window reduces. Therefore ssthresh and
cwnd are set to max (unacknowledged data/2,2_MSS) &
ssthresh+3MSS.TCP Vegas delivers more packets and
TCP Tahoe and Reno relatively send same amount of
Packets. The reason of delivering the more packets for
TCP Vegas is that TCP Vegas retransmits the lost packets
after receiving the 2 duplicate acknowledgements whereas
other TCP Tahoe, Reno, New Reno, SACK and
Westwood retransmit the segments after receiving the
three duplicate acknowledgments but in some cases third
(dupack) takes either long time or does not receive third
dupack and timeout expires. It is advantage of Vegas over
above TCP variants because TCP Vegas mostly
retransmits the lost segments before Retransmission time
out (RTO). The other reason is that TCP Vegas does not
wait for loss to trigger congestion window (cwnd)
reduction. Vegas possesses interesting approach regarding
the congestion because it estimates the level of congestion
before it occurs rather try to avoid it. The level of
congestion is measured on basis of sample RTT and size
of sending window that is also the reason; the Sender
estimates the current throughput against every RTT [11].
TCP Tahoe and Reno have delivered fewer packets than
rest of TCP Variants.
TCP Tahoe faces the problem due to repeat of slow start
phases on each dropped segment, particularly when error
is transient and not constant. In this case congestion
window shrinks and bandwidth cannot fully be utilized.
Fast Recovery algorithm for TCP New Reno can degrade
the performance due to multiple losses of packets during
single window because Fast Recovery algorithm can
manage only single loss per RTT [11, 17]. TCP Tahoe and
Reno Variants have more difficulty to differentiate
between loss and congestion in wireless environment such
as IEEE 802.11.The performance degrading factor for
TCP Reno and Tahoe is also size of congestion window
because these variants cannot send data during the timeout
period, if mainly packets loss occurs [14]. The findings of
our simulation were also validated by researchers in
different papers.
Figure 3: Throughput on different mobility rates at
the maximum speed of 5 m/sec with RW
TCP Reno and Tahoe avoid time outs in case of multiple
consecutive losses occur. The major factor of degrading
the performance for TCP Tahoe has no support of fast
recovery algorithm. This algorithm causes to recover the
lost segments frequently. Figure 4 shows the throughput
performance of TCP Variants at the maximum speed of 10
m/sec. the performance is affected by increasing the speed
and TCP New Reno is severely affected, the reason of
weak performance for TCP New Reno is also aggressive
behavior of fast retransmission algorithm whenever
duplicate acknowledgement (dupack) are received and
high mobility of nodes is available. Due to aggressive
behavior of fast retransmission algorithm, it is difficult to
deliver the packets even partial Acknowledgements (Acks)
are received to sender. Multiple losses due to high
mobility make the weak performance of TCP New Reno
because multiple losses cannot be handled properly and
network becomes more congested and packets start to drop
quickly and this claim is already verified in [4]. Another
performance degrading factor relates to TCP New Reno is
to take one RTT to perceive each packet loss.
Figure 4: Throughput on different mobility rates at the
maximum speed of 10 m/sec with RW
B. (RTT FAIRNESS OF TCP VARIANTS)
We show fairness by sharing the bandwidth among different
TCP variant according Round trip time (RTT) given in figure
5. There are numerous reasons for RTT fairness as one
reason is to attain the equal bandwidth allocation where the
different competing flows may allocate similar bottleneck.
Long RTT consumes more resources than short RTT in
consequence long RTT produces discouraging throughput.
TCP Vegas gets higher throughputs than other TCP Variants
because slow start and congestion recovery algorithms mostly
influence the throughput [12].
Hence slow start and congestion recovery mechanism work in
different way for each TCP Variants because TCP Vegas
depends on difference of expected and actual throughput. The
multiple losses can be retained by avoiding timeouts because
the TCP Vegas retransmit the lost segments after receiving 2
(dupack), as that is reason before timeouts, dropped segments
are retransmitted and better throughput is obtained. Original
feature for TCP Vegas is its congestion detection mechanism
because it shows the problems concerning to fairness.
In congestion avoidance, the congestion detection algorithm
of TCP Vegas verifies every RTT that is benefit of TCP
Vegas over rest of TCP Variants. Moreover TCP Tahoe, TCP
Reno, TCP New Reno and SACK reduce the congestion
windows more than once during the single RTT that is also
reason for unfairness and producing minimum throughputs
where as RTT of Vegas reduces only once during the RTT.
Figure 5: RTT Fairness for TCP Variants at the
speed of maximum 10 m/sec with RW
C. (ENE TO-END DELAY FOR EACH TCP
VARIANTS AT SPEED OF 5 & 10 m/Sec).
End-to-End delay is an average elapsed time for delivery of
individual data packets. All possible delays are included
and caused by routing discovery, transmission at the MAC
layer and queuing at the interface queue, etc but
successfully deliveredpackets are calculated. We show
trend for each TCP Variants in figure 6 and 7. Vegas has
minimum Endto- End delay at the speed of 5m/sec and 10
m/sec whereas TCP Reno and TCP Tahoe have almost
similar maximum End-to-End delay at 5 m/sec. As at the
speed of 10 m/sec, maximum delay has been analyzed for
TCP New Reno. The reason for maximum End-to-End
delay for TCP Reno and TCP Tahoe at the speed of 5 m/sec
is weakness of faster transmission algorithms. Since TCP
Tahoe does not send instant (ACKs) and depends on
commutative (ACKs). Therefore when packet is lost then it
waits for timeout or pipeline is emptied. This causes high
bandwidth delay. TCP Reno behaves like TCP Tahoe
whenever multiple losses occur and multiple losses are
perceived as single segment losses. Another problem
occurs with TCP Reno when the size of window is small;
numbers of duplicates (ACKs) are not detected for fast
retransmission and have to stay for coarse grained timeout.
From other side, TCP New Reno performs weak by
increasing the mobility and takes long End-to-End delay.
Reason is Limitation of retransmitting single lost segment
against per RTT, consequences large delay occurs in
retransmitting the later lost packets in the window.
From other side, if the sender is restricted by the receiver’s
advertised window during recovery time, then the sender is
unable to utilize the existing bandwidth successfully and
takes long End-to-End delay [14]. Minimum End-to-End
delay for TCP Vegas is fairness of retransmission
algorithm when segment is lost that TCP Vegas waits for 2
dupack and retransmit the lost segments before expiry of
timeouts and RTT of Vegas shrinks only once during the
RTT.
Figure 6: Average End-to-End delay for each TCP
Variants at the maximum speed of 5 m/sec with RW
Model
Figure 7: Average End-to-End delay for each TCP
Variants at the maximum speed of 10 m/sec with RW
Model
D. (CONTROL OVERHEAD AT DIFFERENTSPEEDS
WITH RW MODEL)
“This is the ratio between the total numbers of control
packets generated to the total number of data packets
received during the simulation time [16]. Control overhead
contains control packets, which are used to set up a path to
the destination, maintaining and repairing the routes. Control
packets are Route Request (RREQ), Route Response (RREP)
and Route Error (RERR). Figure 8 shows the trend of
control overhead at different mobility ratio. It is very hard to
discover a functional route to destination when speed
increases. Contention and congestion due to the overflowing
behavior of ZRP protocol dominate the effect of the speed.
Whenever speed increases that extra routes are needed in
ZRP. The overhead of control packets increases significantly
as speed increases. Hence more route request segments and
route error are transmitted at the higher. Whenever mobility
ratio and speed increase that more links break, resulting
many control packets are required for route discovery. Due
to increasing of mobility ratio and speed that more segments
travel over non-optimal routes with larger hop counts, which
may be accumulated in a route cache. As a result, these
segments will experience longer End-to-End delay and
causes the creation of many overhead (control packets [13].
ZRP also creates control overhead packets because it often
uses corroded routes due to the large route cache, which
causes frequent segments retransmission and very high delay
times. ZRP is appropriate for networks in which mobile
nodes travel at reasonable speed but not higher. If speed of
nodes is increased, resulting more control overhead (control
packets) is produced. The behavior of routing protocol,
increase in mobility ratio and speed of mobile nodes are
three factors, which creates more control overhead (control
packets).
Figure 8: Control Overhead (Number of control
Packets) at different speeds with RW Model
E. (Number of Broken links due to different
mobility rates)
We show an average broken links for each TCP Variants at
the speed of 5 m/sec and 10 m/sec given in Figure 9. These
broken links are calculated at the 50% mobility. When
MANET nodes want to establish the sessions to obtain
internet services from wired segment of network then
routing protocols start route discovery process. Route
Request packet (RREQ) is broadcasted into network to
obtain any single appropriate route to destination. When
route request packet is reached to destination, in response
route reply packet (RREP) is sent to originator RREQ. If a
link is broken due to mobility and speed of middle nodes, a
route error packet is sent to the destination. Meanwhile
destination finds another route. The process is repeated until
the reply reaches the target. Therefore, destination finds
another route if any error occurs in current route. This
process causes delay in packet delivery. The high mobility
and speed makes more broken links due to discovery of
route. The high mobility and speed continuously change the
direction of node, inconsequence more links break. Due to
increase of speed, topology changes rapidly and more links
are broken particularity in ZRP when more connections are
established between the nodes.
Figure 9: Broken links at different speeds with RW
Model
VIII. DISCUSSION OF RESULTS
We have simulated TCP Variants in ns-2 over Hybrid
network, which consists of MANET, Wired & Wireless
segments of network. We have increased mobility in
MANET segment of network by using ZRP routing
protocol and analyzed that by increasing the mobility and
speed the performance of each TCP Variants gradually
decreases. Multiple routes obviously give benefits but
creates disadvantage due to high mobility. In larger
networks, the source routing principle can also generate a
trouble. It has been observed that TCP Vegas performed
better than other TCP Variants. It produces healthy
throughput, better in-order delivery of data, minimum End-
to-End delay, and good RTT at different mobility ratios and
speeds. The major reasons for degraded performance of
TCP Reno, Tahoe and New Reno are timeouts, as during
the timeouts period, Variants cannot resend the lost
segments whereas TCP Vegas does not wait for loss to
trigger cwnd decrease and calculate approximately the
current throughput during each [15]. Westwood greatly
miscalculates the existing bandwidth, which is potentially
troublesome for fairness and can lead to starvation of
simultaneous connections that is the reason to produce
lesser throughput than TCP Vegas and presented the work
in the paper Performance evaluation of TCP Westwood+
[17]. Hence Losses in wired network are due to overflow of
buffer at routers. TCP Reno, Tahoe and New Reno have
been designed particularly for wired network and meet the
requirement of IEEE802.11 but their performance become
weak in Hybrid network especially satellite link is involved.
Minimum mobility ratios create less control overhead,
which causes the better performance for each TCP Variants,
which is also proved in our simulation. TCP experiences
most losses in multi hop wireless networks, which are
caused by packet drop at wireless link layer IEEE 802.11.
To improve the performance, new congestion control
flavors have been introduced and various schemes are
included. Explicit congestion Notification (ECN) has been
incorporated to improve the congestion control. If
congestion occurs in network then that intermediate routers
will mark the congestion experience (CE) code point in
header of TCP. This message informs the end host that
network is congested and resulting unnecessary packet
drops can be prevented.
IX. CONCLUSION AND FUTURE WORK
Mobile Adhoc networks (MANET) can be deployed to
many locations without the use of infrastructure support. In
military environment, disaster situation, scattered
educational institutions need such networks to route data
packets through dynamically mobile nodes. MANET is
better choice for these extremely mobile and dynamic
applications, which are not supported by centralized
administration. If internet services are required that
MANET is better solution in anywhere to integrate with
wired network to construct as Hybrid network in order to
obtain an internet facility.
To investigate the performance of different transmission
control protocols, we have done simulation in ns-2 by using
Random Waypoint Mobility model and analyzed different
metrics. We have particularly focused on MANET & wired
portions of network to investigate the performance of TCP
Variants. The minimum effect of mobility has been analyzed
on TCP Westwood and reasons are already discussed in
detail but it delivers lesser segments than TCP Vegas and
SACK whereas TCP Vegas has better throughput, minimum
End-to-End delay, better In-order delivery of data and
improved RTT. TCP SACK also performs better and does
not loss many segments because sender is informed which
segment has been received. TCP SACK uses SACK blocks
at receiver side to indicate the contiguous block of data
successfully received. The sender can find out through
SACK blocks which segments are lost, as this is the reason
to control the loss of segments frequently. TCP Reno, TCP
New Reno and Tahoe degrade the throughput in high
mobility ratios and take more End-to-End delay time as
compare to other TCP variants and reasons are already
illustrated. In future, we will analyze and evaluate TCP
Variants in Hybrid network with respect to different mobility
models including social network model, Random Walk
Mobility Model, Random Direction Mobility Model, City
Section Mobility Model etc. We would study under utilized
and congested network conditions by using maximizing
traffic flows. We would also analyze multihoming issues in
future. Finally we suggest if the features of TCP Vegas and
TCP Westwood are combined that new Variant could be
better from mobility point of view in MANET and mixed
environments.
REFERENCES
[1] Razaque. A, Shahzad. K, Qadir. M.A, “Performance
Evaluation of TCP Variants in mobility based Anchor
point node Hybrid Network”, IEEE, ACM & SIGWEB,
CNSR2008. In Proceedings of Sixth Annual Conference on
Communication Networks and Services Research
(CNSR2008) IEEE,ACM & SIGWEB, Halifax, Nova
Scotia, CANADA, pp- ,270-277, May 5 - 8, 2008.
[2] Fernando Tapia, “TCP over mobile ad hoc networks”,
thesis-2004.
[3] Razaque.A,Qadir.M.A, Shahzad.K, Osama Bin
Nazeer, Hammad Khalid, “Device Discovery with
qualitative and quantitative Analysis of TCP Variants
under mobility Based Anchor Point Node Hybrid
Network”, In Proceedings of IADIS International
Conference Wireless Applications and Computing,
Amsterdam, Netherlands. 22 - 24 July 2008.
[4] Qiang Ni,Thierry Turletti and Wei Fu,“ Simulation-
based Analysis of TCP Behavior over Hybrid Wireless
and Wired Networks”, Project RNRT-VTHD++ , France,
October 2002.
[5] A.O. Oluwatope., A. B. Obabire., and G. A.
Aderounmet, ”End-to-End Performance Evaluation of
Selected TCP Variants across a Hybrid Wireless
Network”, In Information Universe Transactions on
Issues in Informing Science and Information
Technology, Vol. 3, pp 479-487, 2006.
[6] Camp, T., Boleng, J., Davies, V,A survey of mobility
models for ad hoc network research. Wireless
Communication and Mobile Computing, Special Issue
on Mobile Ad Hoc Networking: Research, Trends and
Applications, 2(5): 438-502,2002.
[7] Musolesi, M., Hailes, S., Mascolo, C,” An Ad-Hoc
Mobility Model Founded on Social Network Theory”,
Proceedings of the 7th ACM International Symposium
on Modeling, Analysis and Simulation of Wireless and
Mobile Systems (ACM MSWiM’04). Venice, Italy,
p.20-24,2004.
[8] Anthony Lo, Jinglong Zhou, Martin Jacobsson and
Ignas Niemegeers, “ns-2 Models for Simulating a
Novel Beyond 3G Cellular Multihop Network”, ACM
International Conference Proceeding Series; Vol. 202,
Article No. 7, Pisa, Italy, October 10, 2006.
[9] Bai, F., Sadagopan, N., Helmy, A., 2003,” Important: A
Framework to Systematically Analyze the Impact of
Mobility on Performance of Routing Protocols for Ad
Hoc Networks”, Proceedings of the 22nd Annual Joint
Conference of the IEEE Computer and
Communications Societies (INFOCOM’03), 2:825-
835.
[10] Tracy Camp, Jeff Boleng, Brad Williams, Lucas
Wilcox, William Navidi,” Performance Comparison of
Two Location Based Routing Protocols for Ad Hoc
Networks”, Colorado School of Mines, NSF Grants
ANI-9996156 and ANI- 0073699,2003.
[11] Vassilis Tsaoussidis and Ibrahim Matta, “Open Issues
on TCP for Mobile Computing”, NSF grant Career
ANI-0096045, USA, 2002.
[12] Sangtae Ha, Yusung Kim, Long Le, Injong Rhee Lisong
Xu,” A Step toward Realistic Performance Evaluation of
High-Speed TCP Variants”, North Carolina State
University, 2006.
[13]David Oliver jorg, Torsten Braun,” Performance
Comparison Of MANET Routing Protocols In Different
Network Sizes”, computer Science Project, University
of Berne ,Switzerland, 2003.
[14]Kevin Fall and Sally Floyd,” Simulation-based
Comparisons of Tahoe, Reno, and SACK TCP”, U.S.
Department of Energy under Contract No. DE-AC03-
76SF0009, 1996.
[15] Hassan Mahbub, Jain Raj,”High Performance TCP/IP
Networking Concepts, Issues, and Solutions (Book)”,
chapter 11, pp-264-279, 2003.
[14] Dongkyun Kim, Hanseok Bae, Jeomki Song, “Analysis
of the Interaction between TCP Variants and Routing
Protocols in MANETs”, p1, 2005
[15]. Dongkyun Kim, Juan-Carlos Cano , P. Manzoni , and
C-K. Toh ,” A comparison of the performance of TCP-
Reno and TCP-Vegas over MANETs”, pp1-2.
[16] O. J. Oyedapo and D. Ngwenya, “Evaluation of TCP-
Variants Performances in an Ad-Hoc Mobile Network”,
pp1,2007.