Rethinking transport protocols for the Internet

Networks like the Internet only provide best-effort packet forwarding service. Transport protocols provide and end-to-end data delivery service to applications and offer a range of features:

• Reliability ensures that packet arrives despite losses in the network.

• Congestion control aims at reducing the transmission rate in case of network congestion to avoid further packet loss.

The two most widely used transport protocols on the Internet are TCP and UDP. TCP provides a reliable and congestion controlled service to applications. It is widely used for most applications requiring reliability (web, email, etc).

However many applications like multimedia video and voice do not require reliability, and for those, UDP was designed. Unfortunately, UDP does not provide any congestion control and is seen as a threat to the Internet stability. For this reason, the IETF is currently exploring new avenues for transport protocols to replace UDP.

In this context, NICTA researchers are rethinking the transport protocol design.

The current transport protocols are all designed for fixed wired networks. Many assumptions used in this design no longer apply for network technologies such as wireless, mesh networks, and ad-hoc networks. The limitations become even more amplified when considering advanced capabilities such as network service guarantees (QoS) and mobility. The Chameleon Project addresses these limitations by:

• Providing a reliable transport protocols that outperforms TCP on wireless networks
• Developing the first congestion-controlled transport protocol to fully use network bandwidth reservation
• Developing a solution better suited for high-delay links (e.g. satellite)
• Optimizing the design for CPU-constrained devices

RESEARCH AGENDA

Congestion Control effectively limits the amount of traffic injected in the network during period congestion. As a result, it also minimizes packet loss. An established design requirement for congestion control algorithm is that all sources affected react in a same way.

In the Internet, this is also known as "TCP-friendliness" In one type of congestion control algorithm, known as rate-based congestion control, the sender adjusts the transmission rate based on network measurements. TCP-Friendly Rate Control (TFRC) was standardized by the IETF as the current preferred way to realize rate-based congestion control in the Internet.

Reliability is typically achieved using retransmission of lost packets. There are various ways for the receiver to request retransmission from the sender. Currently the most efficient method is Selective Acknowledgment (SACK).

Flow control is concerned with the "impedance mismatch" between the sender and receiver end-hosts performance. Flow control slows down the transmission to adapt the sender to a slower receiver. The classical method is called window-based flow control, where the sender only sends data if there is sufficient buffer space at the receiver.

THE CHAMELEON APPROACH

As operators want to differentiate their networking services by providing throughput guarantees, the current transport protocols cannot take advantage of that. For this reason, we have developed gTRFC, a modification to the TFRC algorithm which enables applications to take full advantage of network resources reservations. gTRFC is currently under consideration by the IETF TSVWG working group.

TFRC, like TCP, exhibits poor performance over large bandwidth-delay product (BDP) links such as satellite links. Chameleon explores modifications to the TFRC algorithm to obtain a homogeneous performance over large BDP ranges. Our current approach adapts the frequency of TFRC feedback messages.

PDA and other CPU-constrained mobile devices used as receivers are negatively impacted by the use of TFRC due to the complexity of the algorithm. Chameleon includes a mechanism to shift the computation from the receiver to the sender side. This has the additional benefit to solve a security problem identified in the TFRC standard, known as the "selfish receiver problem".

Status:

We have already implemented the Sender-based version and the QoS-aware version of the the chameleon protocol in a Java Framework. This framework is available on demand, just send an email to guillaume.jourjon@nicta.com.au.  

The QoS-aware version of the protocol integrates the reliable and unreliable profile of the protocol.

The Flow-Control mechanism is also integrated in the Java prototype and in a modified version of ns-2 TFRC.

We are currently using a discrete event software (Scilab) similar to Matlab in order to optimize the protocol for wireless environment. In this software, we have reproduced a TFRC congestion control and implemented Markov Models of wireless loss patterns. The aim of this on-going work is too adapt the TFRC congestion control to this kind of network and therefore to able a differentiation of wired and wireless loss events.

Bibliography:

 Journal:

• Guillaume Jourjon, Emmanuel Lochin and Patrick Senac, Design, Implementation and Evaluation of a QoS-aware Transport Protocol, Elsevier Computer Communications, Volume 31, Issue 9,pp 1713-1722
• Guillaume Jourjon, Emmanuel Lochin and Laurent Dairaine Optimization of Loss History Initialization, IEEE Communications Letters, Volume 11, Number 3, March 2007, pp 276-278
• Emmanuel Lochin, Laurent Dairaine, Guillaume Jourjon gTFRC, a TCP Friendly QoS-aware Rate Control for Diffserv Assured Service Springer Telecommunication Systems Journal, 10.1007/s11235-006-9004-2, ISSN : 1018-4864 (Print) 1572-9451 (Online), Volume 33, Numbers 1-3 / December, 2006, pp 3-21,

Conference:

• Golam Sarwar, Emmanuel Lochin, Roksana Boreli, Experimental performance of DCCP over live satellite and long range wireless links, 7th International Symposium on Communications and Information Technologies (IEEE ISCIT 2007), Sydney, Australia, October 2007
  
• Guillaume Jourjon, Emmanuel Lochin and Patrick Senac Towards sender-based TFRC, Proceedings of IEEE International Conference on Communications 2007 (IEEE ICC 2007), Glasgow, UK, 24-27 June 2007, Page(s): 1588-1593 Best paper award of the Multimedia Communications & Home Services Symposium of ICC 2007
• Guillaume Jourjon and Emmanuel Lochin and Patrick Senac, Towards a Versatile Transport Protocol CONext 2006 in cooperation with ACM Sigcomm (poster)
• Guillaume Jourjon, Emmanuel Lochin, Laurent Dairaine, Patrick Senac, Tim Moors and Aruna Seneviratne, Implementation and performance analysis of a QoS-aware TFRC mechanism, 14th IEEE ICON 2006 (International Conference on Networking)
• Emmanuel Lochin, Laurent Dairaine and Guillaume Jourjon, gTFRC: a QoS-aware congestion control algorithm, Proc. of the 5th International Conference of Networking.

PhD Thesis:

• Guillaume Jourjon, Towards a Versatile Transport Protocol. Thesis of the University of new South Wales and the University of Toulouse, January 2008.