Abstract: In a mobile ad-hoc network, temporary link failures and route changes occur frequently. With the assumption that all packet losses are due to congestion, TCP performs poorly in such an environment. This study proposes a new mechanism called TASR, TCP-aware source routing which can improve TCP performance in wireless ad-hoc networks. TASR adds a hold state to an existing routing protocol to reduce consecutive timeouts, retransmissions and out-of-ordered packets in TCP. In the simulation study, TASR achieves up to a 60% improvement in performance without requiring any TCP stacks in end systems to be modified.

INTRODUCTION

A mobile ad-hoc network is a network in which a group of mobile computing devices communicate among themselves using wireless radios without the aid of a fixed networking infrastructure. Due to its dynamic property, mobile ad-hoc networks have gained a lot of attention lately as a way of providing continuous network connectivity to mobile computing devices in various areas. However because the topology of networks changes dynamically, route changes frequently and failure to find a valid route promptly would result in a significant drop in performance. When TCP is used as transport protocol such a drop in performance is rather expected because packets sent to an invalid route are all lost. Specifically, TCP performance can suffer due to the following reasons:

So, there have been a lot of research to address the routing problem in mobile ad-hoc networks (Holland and Vaidya, 1999; Chandran et al., 1998; Ahuja et al., 2000; Dyer and Boppana, 2001; Zhang and Wang, 2002). Routes are broken frequently as the movement of mobile terminals gets faster. Frequent route change makes TCP to have multiple packets losses. TCP has only one way to recover from multiple packets losses and that is to expire the retransmission timer but a consecutive timeout makes the RTO (Retransmission Timeout) value exponentially back off (Stevens, 1994) and longer RTO reduces link utilization and decreases throughput dramatically (Vaidya, 2001). So, the researchers think that consecutive timeout is a key factor that affects the TCP performance.

MANET characteristics: The fundamental difference between fixed networks and MANET is that the computers in a MANET are mobile. Due to the mobility of these nodes, there are some characteristics that are only applicable to MANET. Some of the key characteristics are described (Zhang and Wang, 2002).

Dynamic network topologies: Nodes are free to move arbitrarily meaning that the network topology which is typically multi-hop may change randomly and rapidly at unpredictable times.

Bandwidth constrained links: Wireless links have significantly lower capacity than their hardwired counterparts. They are also less reliable due to the nature of signal propagation (Karthik et al., 2009a).

Energy constrained operation: Devices in a mobile network may rely on batteries or other exhaustible means as their power source. For these nodes, the conservation and efficient use of energy may be the most important system design criteria. The MANET characteristics described above imply different assumptions for routing algorithms as the routing protocol must be able to adapt to rapid changes in the network topology.

In this study, we explore a new way to enhance TCP performance by minimizing the consecutive timeout in the case of frequent route changes without modifying TCP stack.

Related work: Recent studies have addressed, the TCP performance problems caused by route failures in a mobile ad-hoc network. Since, TCP assumes that all packet losses are due to network congestion, although the cause of packet losses is route failure, TCP performs congestion control because this behavior is the major reason by which it shows dramatic drop in TCP performance, many studies try to distinguish between route failure and network congestion and thereby improve the performance of the routing protocols (Karthik et al., 2008a).

Chandran et al. (1998) proposed a feedback-based scheme called TCP-Feedback or TCP-F. In this scheme when an intermediate node detects route failure, it explicitly sends a RFN (Route Failure Notification) message to the TCP sender. On receiving the RFN, the TCP sender suspends packet transmissions and freezes all states including the RTO value and CWND (the size of Congestion Window). When an intermediate node learns of a new route to the destination, it sends a RRN (Route Reestablishment Notification) message to the TCP sender which then restores its previous state and resumes transmission. The conclusion of the study was that the average route repair time has a major impact on TCP performance.

TCP aware source routing: Existing routing protocols (Johnson and Maltz, 1996; Perkins and Royer, 1999; Haas and Pearlmane, 2001; Corson and Park, 2001) in ad-hoc networks are designed to reestablish broken routes as soon as possible. However, since there is a delay between when a route is broken and when the source node is informed of the route failure, the packets sent through the broken route will be lost. If UDP is the transport protocol such burst packet losses would not cause a serious problem in performance.

When the source node has alternative routes, the source node keeps sending the packets out through one of them. If the chosen alternative route is again invalid, TCP will suffer a series of consecutive timeouts. If timeouts occur contiguously, the RTO value of TCP’s retransmission timer is exponentially backed off. Since, all packet transmissions are suspended until the timer is expired, link utilization is significantly reduced.

A key observation is that invalid routes cause consecutive timeout. Also note that the chance of invalid routes is quite high because the source node depends on the routing information from its neighbors that are not updated with the route failure. The chance will be higher when mobile terminals move faster in mobile ad hoc networks (Karthik et al., 2008b).

Also note that the congestion control of TCP does not help improve the performance because too many packets are lost and the timer expiration is the only way to start retransmission. Therefore, we need a mechanism that prevents the alternative routes from being invalid, when the source node is informed of a route failure. So, we propose the mechanism called TASR that consists of adding a hold state and refreshing alternative routes in its routing table. Here is how TASR works.

When the source node is informed of a route failure, TASR checks whether the transport protocol is TCP or not. If it is TASR makes the routing protocol transits to the hold state. Hold state is the state in which the routing protocol does not forward any data packets. In the hold state, TASR starts probing alternative route and such probing is performed in parallel. In order to get the fastest path, the destination node should reply to such probing packets as soon as possible and finally when TASR gets the fastest route by n-parallel probing, it escapes from its hold state and starts packet transmission by using that path (Karthik et al., 2009b)

An advantage of TASR is that it does not require the modification of TCP stack. In mobile ad-hoc network, it is accepted that mobile terminals need a new routing protocol because of the unique nature of ad-hoc networks but it is desirable that the transport layer, stays independent of the underlying networks. To the best of the knowledge, all the previous routing protocols in ad-hoc network required TCP stack modification and TASR is the first attempt that achieves no modification.

Design and implementation of TASR: The researchers add a hold state to DSR (Johnson and Maltz, 1996), one of the existing routing protocols in order to implement TASR. The reason we choose DSR is that it is widely accepted in ad-hoc networks, its architecture fits the framework of TASR. Although, we implement TASR by adding a hold state on DSR, we believe that TASR can be applied to any other routing protocol (Karthik et al., 2008c). The DSR protocol is an on-demand routing protocol that is based on the concept of source routing. The protocol consists of two phases: route discovery and route maintenance. When a mobile node wishes to send a packet to a destination node, it first checks its route cache to determine whether it already has a route to the destination node. If the mobile node does not have such a route, it initiates route discovery by broadcasting a route request packet. When either the destination node or an intermediate node that contains a valid path to the destination in its route cache, receives the route request packet, a route reply packet is generated. As the source node receives route reply, route discovery is done and DSR initiates packet transmission.

Route maintenance is accomplished through the use of route error packets. Route error packets are generated at a node when the data link layer encounters a fatal transmission problem. Nodes that receive route error packets remove these invalid paths from their route cache. In order to implement TASR, we define two additional control packets: route probe and reply probe. The route probe packet is used to probe fresh paths when the source node receives a route error packet and reply probe packet is used to reply to the route probe packet. TASR consists of probing routes in n-parallel and adding a hold state.

When DSR reports route error, TASR does not immediately select an alternative path in the route cache but transits to a hold state and start parallel probing. The probing process is as follows: TASR selects n paths from the route cache (we select, heuristic value). TASR sends a route probe message to each path. Destination nodes which have received a route probe, send reply probe messages using reverse paths. When the first reply probe message is received, TASR releases its hold state and resumes packet transmission using that path. It is because the path that has the shortest round trip time is the best path at that given moment. TASR has two ways to escape from the hold state and return to the DSR state. Normally, it transit to hold state→select state→DSR state. However if all ROUTE probe or reply probe messages are lost in immediate nodes, TASR may wait in the hold state infinitely. So we introduce the TASR timer that is used to back to the DSR state. This TASR timer uses timeout value (TTO) be set to:

(1)

where, α is a delay variance factor with a recommended value of 2. Since, TCP’s SRTT adapt itself to mobile ad-hoc networks slowly in a way (Kim and Noble, 2001) we add α, a delay variance factor to RTO. The transition diagram of TASR is shown in Fig. 1. The simulation study is done in the NS2 network simulator (Fall and Varadhan, 1997). NS-2 is a discrete event simulator that was developed as part of the VINT project at the Lawrence Berkeley. National Laboratory. The extensions implemented by the CMU Monarch project enable it to simulate mobile nodes connected by wireless network interfaces. We extended the NS-2 DSR protocol implementation to include TASR. All results are based on a network configuration consisting of TCP-Reno over IP on an IEEE 802.11 (IEEE, 1999) MAC layer.

The network model consists of 30 nodes in a 1500x300 m flat, rectangular area. Each node uses a wireless channel model with a transmission range of 250 m. The nodes move according to the random way point mobility model. The researches measured the throughput of TCP with and without TASR, varying the mean speed from 2-30 m sec^-1 and we got the count of the retransmission timer’s backoff in each case. Also, we measured the sum of each mobile node’s through put in case the network has 5 and 10 TCP sessions. The results are shown in Fig. 2.

Fig. 1:	The transition diagram of TASR

Fig. 2:	Comparison of TCP performance based on routing protocol (DSR,TASR)

Figure 2 shows the throughput of TASR compared to DSR. Significant improvements in throughput can be observed in the best case (e.g., when the mean speed is 30 m sec^-1). The improvement in throughput is largely due to avoiding consecutive timeouts. We can see that the throughput gain increases as the mean speed increases. It is because as the mobility of nodes in mobile ad-hoc network increases, the probability of route failures gets higher. In the current design, TASR initiates only when the protocol of the transport layer is TCP and the routing protocol of the source node received a route errror message.

CONCLUSION

Normal TCP performs poorly in mobile ad-hoc networks because of frequent route changes. In the scheme, TASR does not send out packets until it discovers a reliable route. By holding the state of routing protocol, TASR reduces consecutive timeouts, retransmissions and out-of-ordered packets in TCP. This protocol achieves up to a 60% improvement in performance compared with DSR.

Also, it shows more outstanding improvements in performance as the mobility of mobile terminal increases. We also, experimented with UDP and got a similar result as TCP. In addition, TASR enhances TCP performance just by modifying the routing protocol without any modification of TCP.

ACKNOWLEDGEMENTS

The researchers would like to thank the Director cum Secretary, Correspondent, Principal, SNS College of Technology, Coimbatore for their motivation and constant encouragement.

The researchers would like to thank the Faculty Members of Department of Computer Science and Engineering for critical review of this manuscript and for his valuable input and fruitful discussions.

Also, he takes privilege in extending gratitude to his family members and friends who rendered their support throughout this research.