Mobile network is a part of a global network, which provides access to networks for digital telephony, file downloading, web browsing, as well as location based services (Hassan et al, 2000). In an environment that is prone to errors, the transport protocol design is not simple to use with mobile wireless as in wired network. The Transport Control Protocol is a widely used transport protocol in the Internet Protocol suite. As it was not designed for wireless networks, therefore, it cannot adapt well to a mobile wireless environment, which have high user mobility, bandwidth change, sudden link delay, or high link error rate. To alleviate the situation, the TCP had to undergo enhancement to improve its performance on mobile networks (Goel & Sanghi, 1998). There were proposals to enhance the Selective Acknowledgement and wind scaling of the TCP protocol, and come up with new schemes that could give a better end performance. These schemes included splitting, snooping of the TCP connection among others.
Characteristics of Mobile Networks
Mobile wireless networks have characteristics that distinguish them from wired networks. These characteristics make them have a need for a different transfer layer compared to that used in wired networks. They are characterized with mobility of end hosts, which results in hand-offs, low bandwidth and high link error rate. Mobile networks provide less band width when they are compared to wired links. This is caused by limits on the transmission medium and physical layer design (Goel & Sanghi, 1998). The high rate of error in mobile networks is caused by the following factors. Fading of the multipath occurs when radio frequency signals propagates in multipath between the receiver and the transmitter leading to fluctuations in phase, angle, and amplitude of the signal that reaches the receiver. Doppler shift increases the rate of error, caused by the transmitter and receiver relative velocities. It causes a change in the rate of the signal arriving at the receiver, thereby affecting the reception of the signals. The inter-symbol interference caused by the delay in the arrival of the symbol, which results in the partial withdrawal of the current symbol. Another cause is the decrease in the electromagnetic energy intensity at the receiver, which is caused by long distance between the transmitter and the receiver. Attenuation lowers the ratio between the signal and noise (Goel & Sanghi, 1998).
Mobility of end host is caused by handoffs and blackouts. Blackout is a period of time that a mobile host is detached from the base station. Blackouts can also cause multipath fading, and increase the error rate. A handoff is the transfer process of a mobile host’s transmission from one cell to another, in case, if the mobile host signs off from the home base station, and then logged on in a new station.
Impacts Caused by Mobile Links on TCP Performance
TCP is the major IP traffic transmitter, and it is supported by most network application programs. When mobile networks converge with IP services, various IP service access methods are created. At the same time the end-host computing devices diversity is created. Initially, the TCP was designed for networks that have reliable wired links, as well as stationary hosts. In such conditions, the loss of packets is mainly caused by network congestion (Goel & Sanghi, 1998). The current TCP Reno has two mechanisms that have been put in place to recover the losses. These mechanisms include an algorithm for Fast Retransmit and Recovery, and time out mechanism (Stevens, 2001). During network congestion, the TCP sender utilizes packet loss like an implicit signal, and in this case, the TCP assumes that congestion mainly caused by packet losses. Packet losses are shown by three duplicate ACKs or timeouts, and for the TCP to receive one of these indications, it should continuously increase the network traffic (Hoe, 1996).
The TCP is forced to SLOW-Start phase by the retransmit timer timeouts, which doubles the traffic at every Round Trip Time. On reaching the thresh, the TCP switches to the phase of Constant Avoidance, which increases the traffic linearly at the rate of one Maximum Segment Size per Round Trip Time. When a packet is send and the sender receives three duplicate ACK, the TCP sender quickly retransmits the packet before the packet is lost. This is referred to as Fast Retransmit
The use of Fast Recovery is necessitated by the transmission of lost packet by the TCP sender using Fast Retransmit. This reduces the congestion window size by half.
The use of wireless links in network connection means that handoffs and/or link errors are the main causes of packet losses. The presence of these errors in the network prompts the TCP to register a false network congestion, which invokes unnecessary congestion schemes (Hoe, 1996). The unnecessary congestion control scheme reduces the network through-put, thus increasing end-to-end delays.
There are many scheme proposals to improve the TCP performance in wireless networks. These proposals can be distinguished by the application of two different approaches. The first approach is where the sender knows about the presence of wireless links, and tries to distinguish losses caused by wireless links from losses caused by congestion. This is done to prevent the TCP sender from using congestion control schemes when packet losses are due to wireless errors (Goel & Sanghi, 1998). The second approach is when the sender is not aware of the losses caused by wireless links. The wireless related losses are not recognized the TCP at the host sender, and this means that the TCP at the host sender remains unmodified.
This scheme is effective when it is applied when a congestion window has big packet losses, or when the size of a congestion window is small. This scheme extends Fast Recovery and Fast Retransmit algorithms for TCP flows that have small congestion window with no likelihood of generating three duplicate acknowledgements that triggers Fast Retransmit (Hoe, 1996). When using Limited Transmit scheme, the sender send a packet for unsent packets in the queue of the sender. This is done to respond to each arrival of the first duplicate acknowledgements. Half of a server’s retransmissions are caused by expiration of the timer of the TCP retransmission, close to 25% of such retransmissions can be avoided using Limited Transmit scheme.
When using Selective Acknowledgements, the TCP sender can be informed of the packets that need retransmission in the Round Trip Time after the loss event. Therefore, Selective Acknowledgement allows the sender to recover from the losses of data that is within the loss detection of one RTT. The Selective Acknowledgement, Fast Recovery and Fast Retransmit are able to recover from packet losses, and are able to decrease congestion window in order to avoid further congestion (Hoe, 1996). However, this behavior can result in throughout degradation when there is no packet related to packet losses. This is because the loss of packets is assumed to be an indicator of congestion. When the same assumption is made in the case of wireless network, where major losses are caused by link errors and not congestion, then these will affect the ability of the TCP to find the available bandwidth.
This scheme makes the congestion window show a different behavior when there are corruption losses or congestion losses caused by hand-offs and link errors. Hence, this helps to distinguish the two categories of losses.
Wireless Unaware TCP
This approach makes use of the assumption that most wireless segments connected to the global Internet are near the local end users, and therefore, the problem of packet losses should be solved through local means, and the TCP is independent of the characteristics of individual links in the network (Goel & Sanghi, 1998). The following are the schemes in this approach.
- Delayed Duplicate Acknowledgements and Snoop
The snoop scheme uses the assumption that the mobile wireless link is the TCP connection’s last hop. It also introduces a snoop agent, a module at the base station. A snoop agent caches TCP packets sent across the link, however, the accepted packets are yet to be acknowledged by the receiver. If the agent is cached, it transmits the lost packet, and suppresses the Duplicate Acknowledgements for the lost TCP packets. The retransmission of lost packets is done locally to avoid unnecessary fast retransmissions, as well as the sender’s congestion controls (Goel & Sanghi, 1998). However, this scheme requires a base station in order to retain the state information, as well as cache the TCP packets that are unacknowledged, resulting in scalability issues.
The Delayed Duplicate scheme tries to imitate Snoop’s behavior through the use of link-layer retransmissions. However, this scheme attempts to decrease interference between link-layer transmissions and TCP-layer transmissions, this is done by delaying the 3rd packet and the duplicate of the subsequent packets at an interval of d. The arrival of out-of-order packets at the receiver prompts the receiver to respond by sending duplicate packets for the first packets that are out-of-order (Goel & Sanghi, 1998).
This is a scheme that breaks the wireless mobile host and fixed wired network connection into two connections. The first connection is between a base station and a fixed host and the second connection is between a wireless host and a base station. A base station receives the first data transmitted to the wireless host, and after the base station receives the data, it sends the fixed host an acknowledgement, and then the received data is send to the wireless host. This scheme does not need TCP during the wireless host and the base station communication. Instead, the communication uses a specialized and optimized protocol for mobile applications, unreliable wireless medium as well as low speed medium. This indirection shields wired networks from wireless network uncertainties. However, sometimes the Indirect TCP violates the current TCP acknowledgement mechanism. This happens because the data packets acknowledgements would first reach the source before going to the wireless host.
This scheme was a cellular network proposal for supporting frequent hand-offs and high bandwidth. The scheme can be seen like a three-level hierarchy. Wireless host that communicate with wireless stations in each cell belong to the lowest level. However, a host supervisor at the second level, (Brown & Singh, 1997) runs some of the wireless stations.
The host supervisors are connected to high-speed wired networks at high levels, and take care of most routings, as well as other detailed protocol for mobile users. The M-TCP scheme is used in the communication between mobile stations and mobile hosts on receiving data. A mobile station sends it to a wireless host. However, a mobile station directs the acknowledgement to the TCP sender up to the time it obtains the mobile host acknowledgement (Brown & Singh, 1997). In cases of data losses by the mobile host or a handoff, the mobile station transmits the deferred acknowledgement, and shows a zero sized window, leading to a persist state. When this happens, all timers are frozen up to the time when the mobile host gains connection again. The algorithm gives a solution to frequent and periodic disconnection (Brown & Singh, 1997).
- Freeze- TCP
This scheme is designed to take care of hand-off disconnections, it helps a mobile host in monitoring signal strengths, predict any temporary disconnections and detect an impending handoff. This scheme modifies the TCP algorithm of the mobile host to prevent the base station from transmitting packets during hand-offs (Goel & Sanghi, 1998). During a hand-off, the mobile host shows a zero window size receiver to compel the TPC sender to freeze, and prevent it from showing its congestion window size.
The new transport layer protocol for mobile networks was developed in the late 1990s. This design of transport layer has many applications. The design is SCTP, and it has captured much strength displayed by TCP, which include retransmission, error detection and window based congestion control. This has led to its success in mobile networks and Internet growth as a whole. This design also incorporates new features that TCP does not have, and these features include multi-streaming and multi-homing.
In this case, both A and B are endpoints with interfaces attached to the SCTP association. The End points are joined using two types of links, the ATM at the bottom and the satellite at the top. One address is assigned as the primary and the second is utilized as a backup if the primary address experiences a failure or if there is a request by the upper layer application to use the backup. Lost packets can be retransmitted over secondary address. The multi-homed feature of the SCTP scheme is more useful in environments that need a high availability of applications. This feature can also increase the rate of recovery from situations associated with link failure without interrupting data transfer (Stevens, 2001).
Data is sequenced within a stream, and in case of loss of a segment from a certain stream, subsequent segments from that stream are stored in the stream buffer in the receiver until retransmission of the lost segment from the source is achieved. However, data coming from other streams are allowed to pass to the application in the upper layer. This is meant to avoid line blocking as in the case of TCP, where all data coming from the upper layer application is carried by a single stream. This means that the HOL effect is only within the boundaries of individual streams. However, the entire association is not affected.
Web browsing is an example of SCTP multi-streaming application. A HTML page is divided into five objects, which include plain text, two images, an Active control and a java applet. The SCTP uses multi-streaming feature to increase the rate of transmission of HTML pages, and the transmission of each object using a separate stream that can eliminate the HOL effect between objects (Caro et al, 2003). Therefore, in case of any object loss during transfer, the other objects are delivered to the upper layer of the Web browser, and the lost object is retransmitted from the server on the Web. This gives users a better response time when they are opening the SCTP association for a specific HTML page.
Congestion control using SCTP scheme is based on the TCP control scheme that is window based and rate-adaptive. This control scheme ensure that SCTP reduces the rate of sending data when there is a network congestion, as well as prevent congestion collapse experienced in a shared network (Caro et al, 2003). This scheme ensures reliability in transmission, and detects corrupt, reordered and lost packets. Reliability is provided through the retransmission of corrupt and lost packets.
Buffer-limited user terminal and Bandwidth-limited channels are characteristics of mobile wireless environment. When there are constraints in the receiver buffer, it is advisable to use multi-streaming and not a single stream because of the higher goodput associated with multi-streaming (Caro et al, 2003).
Wireless mobile networks experience frequent delay spikes when they are compared to wired networks. This is a sudden increase in RTT that suddenly drops to its previous value. It is caused by mobile hand-off among other factors, and it results in Spurious Fast Retransmission and Spurious Timeout, which can lead to performance penalty in TCP (Ludwig & Katz, 2000).
This is a proposal by IETF for seamless mobile computing, the use of SCTP in Mobile IP increases the probability of wireless channel’s packet losses that might lead to a back off of transport protocol data transmission rate. SCTP is used together with Selective Acknowledgement in order to bring out a more robust reaction when multiple losses occur in a single window (Caro et al, 2003).
According to research by the University of Oklahoma, a TraSH scheme is the best to use in mobile handover, and with this scheme, handover can be accomplished without on the IP infrastructure (Fu et al, 2003). An example is shown in the figure below
This scheme is different from that used in Mobile IP, where the TraSH requires no Foreign or Home agents, and requires a location manager for use by the CN in the location of the Mobile Handover current position (Fu et al, 2003).
This is a scheme that was studied by Ye and others, in their study. They investigated the RTS threshold, and according to their findings, an exchange of control frames known as RTS/CTS sequence was required in transmitting data large size data frames because a large RTS threshold would lead to a high collision rate, and a small value would lead to a high signaling cost. The researchers have also shown that the degradation of the number of hops occurring between the receiver and the sender increases caused by the SCTP association. This mainly caused by exposed and hidden nodes problems (Ye et al, 2002). Results from the simulation also shows that a hope count that is less than three, low RTS threshold, reduces the collision rate between Selective Acknowledgement data packets and the RTS for DATA packets. Researchers also highlight that the SCTP sender might experience an idling period, caused by a Small Window Syndrome. Therefore, in this case, the sender lacks enough DupACKS to cause a fast retransmission (Stewart & Metz, 2001). However, this setback can be resolved by assuming that MAC layer collision is the main cause of data loss, and transmitting data at a lowest unreceived TSN during the period of idleness. This move can also increase the rate of error recovery. However, it is not valid when transmitting more data in a congested network (Stevens, 2001).
Transport layer in mobile network is a network that is unique from wired network due to the issues of low bandwidth, and congestion among others. Attempts to solve these issues have led to the development of different design schemes such as the improved schemes of TCP transfer protocol and the most recent is SCTP, which has more advanced features such as multi-homing and multi-streaming (Stewart & Metz, 2001). The research in transfer layer in mobile networks has strengthened mobile networks and improved users’ experience. However, more research should be done to come up with more stable schemes that can function well during high data transmissions.