TCP Extensions for Asymmetrical Acess Network

Citation

Teo, Boon Cheng (2002) TCP Extensions for Asymmetrical Acess Network. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

With the increase popularity in multimedia applications over the Internet, the demand for broadband infrastructure is increasing. Currently, the main path to the Internet from home is through the dialup access network, which can only achieve up to 56Kbps. At this rate, the access network has become bottleneck for broadband and multimedia applications. Various technology has been proposed to increase the data rate at the access level. Due to the fact that traffic at the access network is asymmetrical in nature, several high speed asymmetrical access network are proposed, I.e. its downstream bandwidth is significantly higher than that of upstream. Example of asymmetrical access network includes the Asymmetrical Digital Subscriber Line (ADSL), Direct Satellite Broadcast system (DSB) and cable modem network. Among all, ADSL is the most popular and simplest to deploy. TCP, a connection oriented protocol, ensure end-to.end transfer reliability. This is possible with the acknowledgement strategy, which requires an acknowledgement (ack) to be returned for each data packet transmitted. However, this feedbackcontrol mechanism using ack works well only with symmetric network. In asymmetrical network, the returned ack will suffer from the delay due to the low reverse bandwidth. The forward sending rate, which depends on ack arrival, will in turn be reduced. This will result in low utilization of the high speed ( downstream) channel and therefore throughput degradation. With the increase in deployment of asymmetrical network in the future, TCP needs to be extended to target for both asymmetrical and symmetrical network in order to maintain its performance level. We proposed two extensions to TCP: (1) adaptive accumulative acknowledgement scheme and (2) congestion detection mechanism. Unlike some of the proposals that involve the network layer to implement their solution, our solution features end-to-end interoperability to fulfill the OSI layering concept. The first extension features adaptive ack frequency reduction. Frequency of ack can be reduced by sending an ack for more than one data packet received. This is possible because acks are accumulative. By reducing ack frequency, the traffic load of the reverse channel can be relieved significantly, thereby increasing the forward channel utilization. By adaptive, we refer to the ability for the ack reduction factor to synchronize with the number of data packets sent in a batch. With this synchronization, the ack reduction factor varies according to the connection state, be it at the beginning ( slow start), during loss recovery or during congestion avoidance. This is a significant improvement over other ack frequency reduction solutions proposed previously where a fixed reduction factor is used. Congestion detection mechanism is another new idea that we have introduced into TCP. This mechanism enables TCP to detect the relative delays experienced by consecutive data packets in the forward link as well as that between consecutive acks in the reverse link. This is possible by just a simple manipulation on the TCP timestamp values collected in each round trip. As the delay in forward and reverse link is distinguishable, TCP is able to differentiate between congestion in the forward channel from the one in the reverse channel. With existing timestamp option used to evaluate the round trip time, TCP can be aware of congestion in the connection but cannot identify the congested path.

Item Type: Thesis (Masters)
Divisions: Faculty of Engineering (FOE)
Depositing User: Mr Shaharom Nizam Mohamed
Date Deposited: 02 Dec 2009 07:40
Last Modified: 03 Dec 2009 01:36
URII: http://shdl.mmu.edu.my/id/eprint/26

Downloads

Downloads per month over past year

View ItemEdit (login required)