Enhancement of TCP Performance Over  
AD-HOC Networks Using PEP  
 
S. Kulkarni, N. Thanthry, and R. Pendse 
 
Department of Electrical and Computer Engineering, College of Engineering 
 
 
1. Introduction 
       In the past decade or so, wireless devices like PDAs, 
laptops, etc, have invaded the market resulting in an 
increased demand for connectivity with no heed paid to 
physical location. Hence the idea of “mobile ad-hoc 
networks” a.k.a MANETs has caught the fancy of 
researchers around the world. An ad-hoc network is one 
where the formation of the network is dynamic with no 
existing infrastructure in place. When a node enters the 
vicinity of an ad-hoc network, it tries to communicate 
with other nodes and in case it is granted permission, 
becomes a part of the network. The advantage of this kind 
of network is that it can be built and torn down in no time. 
This feature would be very useful in hostile environments 
like war zone, crisis situations, etc. It might also be used 
for business conferencing.  
 
       TCP is a connection- oriented protocol which ensures 
reliable data transmission over the unreliable IP. It is 
responsible for the flow and congestion control within the 
data communication mechanism. TCP is also responsible 
for ensuring that a link’s capacity is shared fairly among 
the connections. It prevents the sender from 
overwhelming the receiver and in case of packet loss 
indicated by non-receipt of an acknowledgement, 
identifies the sequence number of the packet and 
retransmits it. In other words, it dynamically learns the 
delay characteristics of a network and adjusts its operation 
to maximize the throughput without overloading the 
network. Most applications in use like FTP, HTTP make 
use of TCP of TCP/IP suite for their operation. Hence it is 
preferred to implement TCP at the transport layer of an 
ad-hoc network which would facilitate connecting to the 
internet wherever available and support a large gamut of 
applications.  
 
       Research indicates that performance of TCP degrades 
over the wireless networks due to its inability to 
distinguish between packet losses caused due to 
congestion and those caused due to defects in the 
transmission media or due to handoffs. In order to verify the 
performance degradation, an ad-hoc network was built in 
OPNET network simulator. Nodes were randomly placed in 
a square grid of area 5km x 5km. DSR was configured as 
the underlying routing protocol. 12 client/server pairs were 
configured to generate FTP traffic. FTP profile consisted of 
a large file size (1GB) and the application was configured to 
have 50 % command mix, i.e., application was performing 
both download and upload operations equally. 
 
       All nodes were configured to have 11 Mbps bandwidth. 
Nodes were randomly selected to have mobility patterns. 
From the results, it was observed that in case of route 
failure, TCP performance degrades. This performance 
degradation was attributed to the congestion avoidance 
mechanisms used by the traditional TCP as these 
mechanisms reduce the data rate whenever they encounter 
packet drops. It was further observed that the performance 
deterioration is also dependent upon factors like node 
density, area and mobility patterns. In this paper, we 
propose enhancements to the TCP protocol to alleviate the 
performance degradation during events like link failure. 
 
2. Proposed Solutions 
 
       Proposal I:  For a given ad-hoc network containing n 
nodes, data to be sent from a node ‘s’ to a node ‘d’ is routed 
by using either proactive table-driven or reactive routing 
protocols. Consider a situation where route is established 
with the exchange of request and reply packets. The 
concerned data packet is then forwarded along this route. 
After a few hops, the packet reaches an intermediate node 
‘I’ and realizes that the next consecutive link is down. In 
this case the intermediate node sends a feedback to the 
source node informing it of the link failure and requesting it 
not send any further packets on the route. This message is 
piggybacked in the routing table update in case of table 
driven protocols and in hell
 messages in case of reactive protocols. The source now 
makes use of alternate routes present in its route cache to 
forward the remaining packets. However, some packets 
which were already sent along the earlier route are stored 
in the cache of the intermediate node. These packets are 
salvaged by ‘I’. and it attempts to find an alternate route 
to forward these packets. By using this approach, 
retransmissions due to assumed congestion are prevented. 
Also, the sender does not have to retransmit the packets sent 
prior to realizing the occurrence of a link failure since the 
intermediate node takes care of them. 
 143 
 
 
       The success of this approach depends on the trust 
relationship shared between the sender and the 
intermediate node. Even when the intermediate node is 
attempting to recover the already sent packets, the sender 
may resend the same to ensure reliability. The packets 
may arrive out of order at the destination because of the 
different paths followed by the packets.  
 
       Proposal II:  Another approach to improve TCP 
performance is to enable each node within the ad-hoc 
network with PEP ability. This implies that each node is 
capable of sending a proxy reply on behalf of other nodes. 
In this, communication occurs in the following manner: 
source S sends a packet to one of its neighboring nodes 
N1 which sends a proxy reply to S and then tries to look 
for a path to the destination of the packet. In the same 
manner the neighboring node of N1 will send a proxy 
reply to N1 and take over the responsibility of routing the 
packet to its intended destination. This process continues 
till the packet reaches its intended destination D. Thus 
each node receives acknowledgement before timeout and 
hence does not retransmit packets. The biggest challenge 
in this approach is endowing each node with a proxy 
server capability. Even otherwise, there is no guarantee of 
packet delivery. Failure of an intermediate node could 
result in that packet being lost forever. The only way to 
ensure successful transmission of packets is to allow each 
node to be able to view the operations of other nodes 
which, however, would hamper network security. 
 
       Proposal III:  Another approach to improve TCP 
performance also requires us to enable each node within 
the ad-hoc network with PEP ability. Thus each node is 
capable of sending a proxy reply on behalf of other nodes. 
Here, communication occurs in the following manner: 
source S sends a packet to destination D and some where 
along the route a link failure occurs. Say link leading 
from intermediate node N1 to the destination goes down, 
then one of the neighboring nodes of N1 is chosen to send 
a proxy acknowledgement to the source to prevent 
retransmissions and is assigned the responsibility of 
forwarding the packets. All the packets are now routed 
through this proxy PEP agent. The important thing in this 
approach is the criteria used to select the PEP agent. 
Choosing a node with larger number of routes in its route 
cache would result in better performance.  
 
       Proposal IV: Another approach is to assign a static 
node to a group of nodes in an ad-hoc network as a multi 
point relay (MPR). All packets from any source to any 
destination within the network pass through this MPR. 
These MPRs send periodic hello messages to indicate that 
they are connected to their own set of nodes. These MPRs 
prevent retransmissions by sending acknowledgement 
piggybacked in the hello messages to the source indicating 
the successful receipt of a packet intended for any node 
belonging to the set it is connected to. For a destination 
belonging to another MPR, the source MPR will 
communicate with destination MPR and thus ensure reliable 
communication. The only problem with this approach is the 
MPRs can prove to be a single point of bottleneck within 
the network. 
 
3. Conclusions and Future work 
 
       In this project, the author has studied performance 
issues of TCP over ad-hoc networks. Simulation results 
indicate that performance of TCP streams is greatly affected 
by the non- availability of routes. However, the underlying 
TCP manager identifies the performance degradation and 
attributes the cause to congestion and in turn invokes the 
congestion avoidance mechanisms. In this project, the 
author has proposed four different mechanisms that could be 
used to improve the TCP performance in a MANET 
environment.  
 
4. References 
 
1. H. Balakrishnan and R. Katz, “Explicit loss notification and wireless 
web performance,” in IEEE Globecom Internet Mini-Conference, 
Sydney, Oct. 1998. 
2. H. Balakrishnan, V. Padmanabhan, S. Seshan, and R. Katz, “A 
comparison of mechanisms for improving TCP performance over 
wireless links,” in ACM SIGCOMM, Stanford, CA, Aug. 1996. 
3. H. Balakrishnan, V. N. Padmanabhan, and R. H. Katz, “The effects of 
asymmetry on TCP performance,* in Proceedings of the IEEE 
Mobicom‘97, (Budapest, Hungary), pp. 77-89, September 1997. 
4. Analysis of TCP Performance over Mobile Ad Hoc Networks. Gavin 
Holland and Nitin Vaidya Department of Computer Science Texas 
A&M University, College Station, TX 77843-3112. 
5. ATCP: TCP for Mobile Ad Hoc Networks by Jian Liu SUN 
Microsystems, Palo Alto, CA 94303 ;Suresh Singh Department of 
Computer Science Portland State University Portland, OR 97201. 
  
 
 144