A Feedback Based Scheme For Improving TCP

Performance In Ad-Hoc Wireless Networks 

Kartik Chandran, Sudarshan Raghunathan, S. Venkatesan, Ravi Prakash

Computer Science Program
University of Texas at Dallas
Richardson, TX 75083-0688
fchandran,rsud,venky,[email protected]

Abstract
Ad-hoc networks consist of a set of mobile hosts that communicate using wireless
links, without the use of other communication support facilities (such as base stations).
The topology of an ad-hoc network changes due to the movement of mobile hosts, which
may lead to sudden packet losses and delays. Transport protocols like TCP, which have
been built mainly for xed networks, misinterpret this loss as congestion and invoke
congestion control. This leads to unnecessary retransmissions and loss of throughput.
To overcome this problem, we propose a feedback scheme, whereby the source can
distinguish between route failure and network congestion. The main idea is to inform
the source by a Route Failure Notication(RFN) when the route is disrupted allowing
the source to freeze its timers and stop sending packets (new or retransmitted) as
long as the source cannot reach the destination. When the route is re-established,
the source, on being informed through a Route Re-establishment Notication(RRN),
resumes by un-freezing timers and continuing packet transmissions. The simulated
performance of TCP on ad-hoc networks with and without feedback is compared and
reported. We observed that in the event of route failures, as the route re-establishment
time increases, the use of feedback provides signicant gains in throughput as well
as savings in unnecessary packet transmissions. Several further enhancements and
directions for future work are also sketched.
 This work was supported in part by the Texas Advanced Technology Program under Grant No. 9741-052
and by NSF under Grant No. CCR-9796331.

1
1 Introduction
Ad-hoc networks consist of a set of mobile hosts communicating amongst themselves using
wireless links, without the use of any other communication support facilities (such as base
stations). They are also called mobile radio networks or multi-hop wireless networks. Two
mobile hosts (MHs) are said to be within range and said to be neighbors of each other if
each can receive the other's transmission. Every MH behaves in a co-operative fashion by
acting as a router allowing packets destined to other MHs to pass through it. Thus, if MH
h1 needs to send a packet to h2 and h2 is not within the range of h1 , then the packet must
be sent to one of h1's neighbors, say hi, and hi will forward that packet to its neighbor, and
so on, until it reaches the destination, h2. In this example, hi acts as a router.
The topology of an ad-hoc network changes every time a mobile host's movement results
in establishment of new wireless links (mobile host moves within the range of another) or
link disconnections (mobile host moves out of the range of another which was within its
range). The rate of topology change is dependent on the extent of mobility of the hosts and
the transmission range of the hosts. Routes are heavily dependent on the relative location of
MHs. Hence, routes may be repeatedly invalidated in a sporadic and arbitrary fashion due
to the mobility of hosts. The mobility of a single node may aect several routes that pass
through it. For example in Figure 1, the two routes, one from s1 to d1 and the other from
s2 to d2, are changed when MH i moves close to MH h.
Ad-hoc networks have been studied extensively in the context of routing and a num-
ber of routing protocols have been proposed for ad-hoc networks including distance vector
schemes [16], link reversal [6], TORA [15], dynamic source routing [11], routing using a
virtual backbone [7], zone routing [9] and cluster based routing [13].
In this paper, we consider the problem of maintaining reliable stream-oriented end-
to-end communication in ad-hoc networks, similar to that provided by TCP over the
Internet. The end-to-end communication problem has been studied in the context of cellular
wireless networks and a number of modications and extensions to regular TCP for such
networks have been proposed [1, 2, 3, 4, 18, 19]. It is desirable to use TCP directly even
in ad-hoc networks in order to provide seamless portability to applications like le transfer,
email and WWW browsers written using standard TCP libraries. Hence, it is of interest to
2
 h h

c g a g
S1 a j
D2 S1
c
D2

i i
f k
S2 D1 S2 D1
b d b d f

e e

Original Routes between New Routes between

source-destn pairs Source-destn pairs

Figure 1: Eect of Mobility on Routes in Ad-Hoc Networks

study the behavior of TCP in the context of ad-hoc networks and evaluate the eect of the
dynamic topology on TCP's performance. This is done to determine whether it is feasible to
use TCP without substantial changes in the ad-hoc network domain. Our preliminary studies
and results indicate that as a result of frequent and unpredictable route disruptions, TCP's
performance is indeed substantially degraded both in terms of throughput and goodput
(the ratio of useful data received at the destination and the total data transmitted by the
source) [3]. We therefore propose a feedback based scheme for overcoming this problem. Our
simulation experiments show that the use of feedback mechanisms along with TCP can result
in substantial performance improvements.
Section 2 of the paper describes the ad-hoc network model and assumptions. Section
3 discusses the behavior of TCP in ad-hoc networks and Section 4 explains our proposed
approach for improving TCP performance. In Section 5, our simulation model, experiments
and observations are presented and discussed, followed by conclusions and proposed future
extensions in Section 6.

2 Model and Assumptions

For the forthcoming discussions, we make the following assumptions:

3
 1. An ad-hoc network consists of n MHs, each of which is equipped with wireless com-
munication capability. Each MH broadcasts its packets on the wireless channels. No
assumption is made about the medium access protocol.
2. The wireless links are bidirectional. This implies that all the mobile hosts have the
same transmission range. Otherwise, the weaker hosts can receive the transmissions of
stronger hosts but not vice versa, if they are suciently far away.
3. A reliable data link layer protocol is implemented over the unreliable wireless links by
using, say, link level acknowledgement and retransmission. Therefore, we assume that
if a packet cannot be delivered at the link layer, the higher layers will be duly informed.
4. A suitable routing protocol from among those mentioned in Section 1 is implemented to
establish and maintain routes between a source and a destination. We will require the
routing protocol to perform certain actions in addition to forwarding packets, which
may include sending feedback messages to transport entities. The routing protocol
may maintain redundant routes between dierent sources and destinations.
5. When a packet cannot be sent on a link despite the presence of a reliable link layer
protocol, we treat the situation as the failure of the link due to mobility. The routing
protocol is then responsible for adapting and maintaining routes between all sources
and destinations. We assume that when a route failure occurs, a nite time elapses
until the route is restored and communication can be resumed.
6. All packets carry the source and destination id's so that the network layer can identify
the source and destination address of each packet.
A transport connection is made between a source and a destination. The source has a se-
ries of packets that are to be transmitted to the destination. Note that this transport connec-
tion is unidirectional - packets go from source to destination (and acknowledgements go from
the destination to the source). In general, the transport connection may be unidirectional
or bidirectional. For ease of exposition, we will treat the connection to be unidirectional.

4
3 Behavior of TCP in Ad-hoc Networks
TCP is a reliable, stream-oriented transport layer protocol which has been designed for use
over xed networks like the Internet [5]. It has been established that packet error/loss rates
over the Internet due to transmission errors are of the order of 1% [10]. In other words,
packets are rarely lost. Route failures and disruptions are very infrequent as the network
is xed. Therefore, packet loss, which is detected by TCP as a timeout, is interpreted to
be a symptom of congestion in the network. In response, TCP invokes congestion control
mechanisms [10, 12, 17]. In other words, TCP cannot distinguish between congestion on the
one hand and packet loss due to transmission errors or route failures on the other. This
inability of TCP to distinguish between two distinct problems exhibiting the same symptom
will result in performance degradation in ad-hoc networks, as described later in the section.
In an ad-hoc network, packet losses are frequent in the error-prone wireless medium.
However, the eect of these losses can be reduced using reliable link layer protocols. Route
failures, which can occur frequently and unpredictably during the lifetime of a transport
session, depending on the relative motion of MHs in the network, are more dicult to handle.
In general, whenever the mobility of an MH invalidates a route, the re-establishment of the
route by the underlying routing protocol takes a nite amount of time. During this period
of time, no packet can reach the destination through the existing route. This will result in
the queuing and possible loss of packets/acknowledgements, which will be interpreted by the
transport protocol at the source as congestion.
Consequently the source will:
1. Retransmit unacknowledged packets upon timing out.
2. Invoke congestion control mechanisms that include exponential backo of the retrans-
mission timers and immediate shrinking of the window size, thus resulting in reduction
of the transmission rate.
3. Enter a slow start recovery phase to ensure that the congestion has reduced before
resuming packet transmission at the normal rate.
This is undesirable for the following reasons:
5
  When there is no route available, there is no need to retransmit packets that will
anyway not reach the destination.
 Packet retransmission wastes precious MH battery power and scarce bandwidth.
 In the period immediately following the re-establishment of the route, the throughput
will be unnecessarily low as a result of the slow start recovery mechanism even though
there is actually no congestion in the network.

4 TCP-F: A Feedback Based Approach

From the preceding discussion, it is clear that treating route failure as congestion and invok-
ing congestion control is not advisable as congestion control and route failure are disparate
phenomena which have to be handled independently and separately. Therefore, we propose a
scheme by which the source is informed of the route failure so that it does not unnecessarily
invoke congestion control, and it refrains from sending any further packets until the route is
restored. Feedback based schemes for TCP have already been proposed in the form of ex-
plicit congestion notication (ECN) [8] for xed networks and EBSN [3] in cellular networks.
ECN is used in order to hasten the process of congestion detection while in our case we use
feedback to explicitly notify the source of route failure. EBSN is used in cellular networks to
handle transient wireless errors using the reliability of the xed portion of the network. As
we do not have a reliable backbone in case of ad-hoc networks, neither of these methods is
directly applicable in our case. Therefore, we propose a feedback based scheme for handling
route failures in ad-hoc networks termed as TCP-Feedback or TCP-F, which is described
below:
Consider, for simplicity, a single bulk data transfer session, where a source MH is sending
packets to a destination MH. As soon as the network layer at an intermediate MH (hence-
forth referred to as the failure point or FP) detects the disruption of a route due to the
mobility of the next MH along that route, it explicitly sends a Route Failure Notication
(RFN) packet to the source and records this event. Each intermediate node that receives
the RFN packet invalidates that particular route and prevents incoming packets intended

6
 TCP State Machine

RFN From SYN-RECVD

From SYN-SENT

SNOOZE ESTABLISHED

To FIN-WAIT_1

To CLOSE-WAIT

RRN
or Route Failure Timeout

Figure 2: The TCP-F State Machine

for that destination from passing through that route. If the intermediate node knows of
an alternate route to the destination, this alternate route can now be used to support fur-
ther communication and the RFN is discarded. Otherwise, the intermediate node simply
propagates the RFN towards the source.
On receiving the RFN, the source goes into a snooze state and performs the following
steps:
1. It completely stops sending further packets (new or retransmissions).
2. It freezes (i) all of its timers, (ii) the send window of packets (iii) values of other state
variables such as retransmit timer value and window size and (iv) starts a route failure
timer which corresponds to a worst case route reestablishment time. The timeout
value of this timer can be a parameter whose value depends on the underlying routing
protocol.
The source remains in this snooze state until it is notied of the restoration of the route
through a Route Re-establishment Notication (RRN) packet as explained below:
Let one of the intermediate nodes that has previously forwarded an RFN to the source,
learn about a new route to the destination (through a routing update). This intermediate
node then sends an RRN packet to the source (whose identity it had previously stored). All

7
further RRNs received by this intermediate node to the same source are discarded. Any
other node that receives the RRN simply forwards it towards the source.
As soon as the source receives the RRN, it changes to an active state from the snooze
state|it restarts the timers from their frozen values (and does not reset the timers) and
resumes the transmission based on the stored sender window and timeout values. These
steps in eect reduce the eect of TCP's congestion control mechanism when transmission
re-starts. Communication now resumes at the same rate as that before the route failure
occurred. To ensure that the source does not indenitely remain in the snooze state waiting
for an RRN which may be delayed or lost, we use a Route Failure Timer. When the source
receives the rst RFN, it starts this timer. When the Route Failure Timer expires, we
restart the frozen timers (as though an RRN was received) and let TCP's congestion control
mechanism handle the failure.
Figure 1 shows the state machine of TCP-F.
We have conducted simulation experiments based on TCP-F and have compared it with
basic TCP. Our results, which are described in the next section, suggest that TCP-F results
in signicant improvements in throughput as well as goodput.

5 Simulated Performance
To compare the behavior of TCP with that of the proposed TCP-F mechanism, as a prelim-
inary study, we simulated a single unidirectional bulk transfer session.

5.1 Simulation Model

The source sends a continuous stream of data to the destination through the ad-hoc network.
Since we are only interested in studying transport protocol behavior, we view the network
as a black box (containing a xed number of nodes between source and destination) that
emulates the behavior of an ad-hoc network from the viewpoint of a transport entity. The
reconguration of the network (due to the movement of MHs) manifests itself in the form
of route failures and route re-discoveries. The transport entity sees this eect in the form
of sudden packet losses and delays. The actual location of the route failure is a random

8
 Data Packets

SOURCE Ad-Hoc Network Subnet DESTN

Acknowledgements from Destn

RFN/RRN from Failure Point

Figure 3: Simulation Model

parameter and route re-establishment time and frequency of failures are parameters of the
simulation. No assumption is made about the routing protocol used by the underlying
network layer. The simulation model is shown in Figure 3.

5.2 Implementation
Simulation Parameters: For our experiments, we used a xed packet size of 200 bytes
and data rate of 12.8 Kbps (on the wireless channels), which are reasonable values for wireless
networks [3]. The number of hops between the source and destination is set to 10 and the
corresponding window size is 4 KBytes (20 packets). We have assumed that the network in
our simulation does not suer from congestion. Consequently, any packet loss is attributed
to route failure only. The input parameters to the simulation are the failure rate (number
of failures in the total time of simulation) and route re-establishment delay (RRD).
Each run of the simulation is for a period of 100 seconds and is repeated 10 times. The
average of the ten values are reported.
The simulation is event-driven and proceeds as follows: Transmission of a packet leads
to a SEND event at the source which in turn triggers an ACK event and a TIMEOUT event.
The timestamp of the ACK event is calculated as follows: tACK = tSEND + delay based on
number of hops, data rate and packet size + random variance (of up to 10% of total time

9
 1.5 12

1.4
10
1.3

Percentage of retransmitted packets

1.2 8
Throughput (KBytes/s)

1.1
6
1

0.9 4

0.8 4 Failures: TCP

4 Failures: TCP-F
2 4 Failures: TCP
4 Failures: TCP-F
0.7

0.6 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Route re-establishment delay (RRD) (seconds) Route re-establishment Delay (RRD) (seconds)

Figure 4: Throughput and retransmissions for 4 failures with data rate of 12.8 Kbps

to account for packet delays). The timestamp of the TIMEOUT event is based on TCP's
timeout estimation mechanism. Depending on the conditions as described below, one of
the above events (ACK/TIMEOUT) will be valid and the other event is then discarded. In
order to simulate the reconguration of the network due to mobility, route failure events are
periodically generated. The route is re-established after a delay corresponding to the RRD.
Based on the current time instant, location and duration of the failure and the timestamp
of a SEND event, we can determine whether a packet reached the destination successfully
and if an ACK successfully reached the source. Once we have determined which packets are
\lost," their ACK events are then deleted from the event queue.
Over this model, we implemented basic TCP (along with it's retransmission scheme) and
the proposed TCP-F. We then ran the simulation for both cases using dierent values of fail-
ure interval (i.e. number of failures in the period of observation) and route re-establishment
delay (RRD). It was ensured that both TCP and TCP-F were simulated under the same
conditions.

5.3 Observations
Figures 4, 5 and 6 show the behavior of TCP with and without feedback as a function
of RRD. We observe that, for low RRD values, TCP and TCP-F show almost identical
behavior. This is consistent with our intuitive expectations: for low RRD values, fewer
packets will be lost, leading to fewer timeouts and retransmissions (Figure 7).Hence, the

10
 1.5 14

1.4
12
1.3

Percentage of retransmitted packets

1.2 10
Throughput (KBytes/s)

1.1
8

6
0.9

0.8 7 Failures: TCP 4

7 Failures: TCP-F
0.7
2 7 Failures: TCP
0.6 7 Failures: TCP-F

0.5 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Route re-establishment delay (RRD) (seconds) Route re-establishment Delay (RRD) (seconds)

Figure 5: Throughput and retransmissions for 7 failures with data rate of 12.8 Kbps
1.5 16

1.4 14

1.3
Percentage of retransmitted packets

12

1.2
Throughput (KBytes/s)

10
1.1
8
1
6
0.9

4
0.8 10 Failures: TCP
10 Failures: TCP-F

0.7 2 10 Failures: TCP

10 Failures: TCP-F

0.6 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Route re-establishment delay (RRD) (seconds) Route re-establishment Delay (RRD) (seconds)

Figure 6: Throughput and retransmissions for 10 failures with data rate of 12.8 Kbps

extent of timer backo is limited, enabling TCP to recover quickly once the route is restored
(see Figure 8 with RRD = 0.8). The comparable performance of TCP-F suggests that the
feedback does not interfere with TCP during the normal course of operation.
However, as the RRD increases, more packets and acknowledgements are likely to get
lost, due to which more timeouts will occur. For each such timeout in TCP, the retrans-
mission timer's timeout value is increased exponentially. When the route is subsequently
re-established, TCP is unable to resume normal transmission until the retransmission timers
(whose timeout values are likely to be high) expire . This leads to severe throughput degra-
dation immediately after the failure. In contrast, in the case of TCP-F, the receipt of an

11
 800 800
Route re-establishment delay = 0.8 sec. Route re-establishment delay = 2 sec.

700 700

600 600
Sequence number of packet

Sequence number of packet

500 500

400 400

300 300

200 200

100 TCP-F 100 TCP-F

TCP TCP

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 110
time (seconds) time (seconds)

Figure 7: Sequence number of packets sent for 5 failures with data rate of 12.8 Kbps
1000 10000
Route Re-establishment Delay = 0.8 sec. Route Re-establishment Delay = 2 sec.

1000
100
timeout value (seconds)

timeout value (seconds)

100

10

10

1
1
TCP-F
TCP TCP-F
TCP

0.1 0.1
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 110
time (seconds) time (seconds)

Figure 8: Timeout values for 5 failures with data rate of 12.8 Kbps

RFN at the source results in the freezing of its state and reduces unnecessary SENDs as
well as timeouts, thus controlling exponential backo. Therefore, the source can now resume
transmission earlier than in case of TCP, which leads to increased throughput and reduc-
tion in the number of retransmissions. Figures 7 and 8 for an RRD value of 2 seconds and
with 5 failures distributed uniformly during the simulation time, clearly illustrate the above
behavior. From Figure 8, with RRD value of 2 seconds, we observe that the timer values
of basic TCP have increased to a great extent and for an extended period of time. On the
other hand, the timer values for TCP-F resemble short-lived spikes due to which their eect
on packet transmission is negligible.

12
 110 2

100 1.8

1.6
90

Percentage of retransmitted packets

1.4
80
Throughput (Kbps)

1.2
70 4 Failures: TCP
4 Failures: TCP-F
1
60
0.8
50
0.6

40
0.4 4 Failures: TCP
4 Failures: TCP-F
30 0.2

20 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Route re-establishment delay (RRD) (seconds) Route re-establishment Delay (RRD) (seconds)

Figure 9: Throughput and % retransmissions vs. RRD for 4 failures, data rate 128 Kbps

The eect of data rate on performance: On running experiments with dierent data
rates, we found that the improvement in performance of TCP-F over regular TCP increases
with data rate. We see from Figure 9 that there is a much steeper fall in the throughput
curve and a corresponding rise in the retransmissions curve as the value of RRD increases.
Intuitively, for a given time interval, the number of packets passing through the network
increases with data rate. Consequently, if we consider a failure that lasts t seconds, the
number of packets that are lost in time t increases with data rate, which leads to further per-
formance degradation in TCP. Therefore, as data rates in wireless media increase, feedback
will provide even greater benets.

5.4 Discussions
The use of feedback raises some interesting issues that are discussed below.
1. Eect of multiple failures on the same route: It is always possible that multiple route
failures occur independently along dierent links of the route. This is however not
a serious concern in case of TCP-F as the source will then receive an RFN from the
nearest FP and behave accordingly. The communication will then resume only after
the source receives an RRN from that failure point (FP), which means that the route
13
 has been restored.
2. The eect of congestion on the feedback mechanism: It is possible that in a congested
network, the RFN and RRN packets may be lost or delayed. However, this is not a
concern since basic TCP at the source will detect congestion and invoke congestion
control. This behavior is desirable as we are only attempting to distinguish packet loss
due to congestion from that due to route failure; we are not interfering with TCP's
congestion control mechanism in any way.
3. Eect of failure on multiple transport connections: In the current discussion, we have
considered a single transport connection in the simulation analysis. However, in a
real life situation, there may be a number of transport connections that use the same
route/link. Thus, a single route/link failure is likely to aect a number of transport
connections. This raises the following questions:
(a) How do we determine all sources aected by the failure ?
(b) How do we inform these sources?
Note that route failure is detected by the network layer at the failure point, which is
completely unaware of any end-to-end transport connections. Therefore in order to detect
and inform all of these sources, we can use the following approach: Whenever the failure
point receives a packet, it sends an RFN in response to this packet to the source (assuming
that the received packet carries the source and destination addresses). This also ensures that
a source, which is not currently using this route, does not know about the route failures.
We are currently exploring methods to eciently inform all of the sources that use the route
that is disrupted.

6 Conclusions
We have studied the eect of route failures, which are characteristic of ad-hoc mobile net-
works, on basic TCP's performance. In TCP, if the source is not aware of the route failure,
then the source continues to transmit (or re-transmit) packets even when the network is

14
down. This leads to packet loss and performance degradation. Moreover, since packet loss
is interpreted as congestion, TCP invokes congestion recovery algorithms when the route is
re-established, leading to unnecessary throttling of transmission. We have proposed a feed-
back based scheme, in which the failure point noties the source of route failure and route
re-establishment, thus distinguishing route failures from congestion. We have studied the
relative performance of basic TCP and TCP-F and found the results to be very encourag-
ing. As the route re-establishment delay grows, TCP-F performs signicantly better than
TCP. This is attributed to the fact that we are able to reduce the number of unnecessary
packet re-transmissions/timer back-os during the route failure interval. Also, as data rates
of wireless channels increase, TCP-F's advantage over TCP grows.

6.1 Extensions and Future Work

Throughout the paper, we have assumed that packets reaching the failure point are lost
when the next link from the failure point is down. However, this is not so if the intermediate
nodes can buer these packets to a limited extent. In this case, we may do the following:
As the RFN propagates to the source from the failure point, all the intermediate nodes can
temporarily buer subsequent packets. If there is a substantial overlap between the newly
established route and the old route, the RRN message can be used to ush out the buers.
That is, the buered packets may be sent to the destination along the newly established
route. Similarly, on learning of new routes to the destination, the intermediate nodes may
forward the buered packets to the destination, without waiting for an RRN. This buering
scheme has the following advantages:
1. It will save packet retransmissions, and packet ow can resume even before the source
learns about the route re-establishment.
2. Since the buering is staggered across the intermediate hops, the buering overhead
at each node is expected to be low.
Another extension is to study the acknowledgement policies in TCP. When a failure
occurs, a number of packets and/or acknowledgements may be lost while subsequent packets

15
may reach the destination after the route has been re-established. This is likely to result
in gaps in the receiver's window, thus aecting TCP's cumulative acknowledgement scheme.
Therefore it may be worthwhile exploring alternative end-to-end acknowledgement schemes
such as Selective Acknowledgement (SACK) [14] and compare their relative performance
with the performance using cumulative acknowledgement schemes in ad-hoc networks.
We plan to study the mutual eects of feedback and congestion on TCP performance in
greater detail. We are currently working on a more extensive simulation involving multiple
nodes and multiple transport connections to further validate our claims.
Acknowledgement
The authors wish to thank Sridhar Alagar, Ramki Rajagopalan, and Charles Shields for
various comments and discussions.

References
[1] Balakrishnan, H., Seshan, S., Amir, E., and Katz, R., H. Improving TCP/IP
Performance over Wireless Networks. In Proceedings 1st ACM Conf. on Mobile Com-
puting and Networking , November 1995.
[2] Bakre, A., and Badrinath, B.,R. I-TCP: Indirect TCP for Mobile Hosts. In
Proceedings of the International Conference on Distributed Computing Systems October,
1994.
[3] Bakshi, B., S., Krishna, P., Vaidya, N., H., and Pradhan, D.K. Improv-
ing Performance of TCP over Wireless Networks. In Proceedings of the International
Conference on Distributed Computing Systems May, 1997.
[4] Caceres, R., and Iftode, L. Improving the Performance of Reliable Transport
Protocols in Mobile Computing Environments. In Journal on Selected Areas In Com-
munication, Spl Issue on Mobile Computing Networks (1994), IEEE.
[5] Comer, D. Internetworking with TCP/IP. Volume 1 (Second Edition): Principles,
Protocols and Architecture. Prentice-Hall Inc., 1991.
[6] Corson, S., Macker, J., and Batsell, S. Architectural Consid-
erations for Mobile Mesh Networking. Internet DRAFT RFC Version 2.
http://tonnant.itd.nrl.navy.mil/mmnet/mmnetRFC.txt May, 1996.
[7] Das, B., Sivakumar, R., and Bharghavan, V. Routing in Ad-Hoc Networks Using
a Virtual Backbone. To Appear in IEEE IC3N , 1997.
16
 [8] Floyd, S. TCP and Explicit Congestion Notication. ACM Computer Communication
Review, Vol 5, No. 5 October, 1994.
[9] Hass, Z.J. A new routing protocol for the recongurable wireless networks. Proceedings
of International Conference on Universal and Personal Communication, October 1997
(to appear).
[10] Jacobson, V. Congestion Avoidance and Control. In Proceedings of ACM SIGCOMM
August, 1988, pp. 314-329.
[11] Johnson, D., B., and Maltz, D., A. Dynamic Source Routing in Ad-Hoc Wireless
Networks. Tomasz Imielinski and Hank Korth, eds, Mobile Computing Kluwer Academic
Publishers, 1996.
[12] Karn, P., and Patridge, C. Estimating Round Trip Times in Reliable Transport
Protocols. In Proceeding of ACM SIGCOMM August, 1987.
[13] Krishna, P., Chatterjee, M., Vaidya, N., H., and Pradhan, D., K. A Cluster-
based Approach for Routing in Ad-Hoc Networks. 2nd USENIX Symp. on Mobile &
Location-Independent Computing April, 1995.
[14] Mathis, M., Mahdavi, J., Floyd, S., and Romanow, A. TCP Selective Acknowl-
edgement Options. Internet DRAFT RFC 2018 October 1996.
[15] Park, V.,D., and Corson, M., S. A Highly Adaptive Distributed Routing Algorithm
for Mobile Wireless Networks. In Proceedings of IEEE INFOCOM'97 April, 1997.
[16] Perkins, C., E., and Bhagwat, P. Destination Sequenced Distance-Vector Routing
(DSDV) for Mobile Computers. In Proceedings of ACM SIGCOMM Conference on
Communication Architectures, Protocols and Apps. August, 1994, pp. 234-244.
[17] Shenker, S., and Zhang, L. Some Observations on the Dynamics of Congestion
Control Algorithms. ACM Computer Communication Review October, 1990, pp. 30-39.
[18] West, S., and Vaidya, N., H. TCP Enhancements for Heterogeneous Networks.
Technical Report, #97-003, Department of Computer Science, Texas A&M University,
April, 1997.
[19] Yavatkar, R., and Bhagwat, N. Improving End-to-End Performance of TCP over
Mobile Internetworks. In Proceedings of Workshop on Mobile Computing Systems and
Apps. December, 1994.

17