Tailor-made Congestion Control

The broad range of modern Internet applications, ranging from a web browser to highly complex Grid computing Software, calls for flexible methods to use the underlying network infrastructure. Nowadays, flexibility is restricted by the standard congestion control mechanism implemented in the TCP protocol. While there are many proposals for new congestion control mechanisms that are fair towards TCP and advantageous for special applications (e.g., by showing a smoother rate with a datagram oriented instead of a congestion oriented setvice for streaming media), it is currently up to the application programmer to implement thern. During this project, a middleware for tailor-made congestion control will be designed, implemented and tested. This middleware organizes the network infrastructure by choosing and tuning an appropriate congestion control mechanism based an Quality of Service (QoS) requests and a traffic specification from the application; for example, if the traffic caused by the application is known to be interrupted sporadically for short time periods and delay is a less important QoS parameter than bandwidth, it may be better to compensate for missing data by buffering than restarting the congestion control mechanism (as in the normal Gase). Another application may prefer a mechanism that yields a relatively small loss ratio at the cost of large rate fluctuations. In general, such factors are trade-offs; knowledge of the application is required to make an appropriate choice. The new middleware additionally monitors the network performance and provides high-level QoS feedback to the application based an these measurements. lt introduces a new abstraction layer - an interface between the application, where QoS is required, and the network, where efficient resource utilization is the primary goal; thereby, a step towards bridging the gap between the two research domains "Quality of Service" and "Congestion Control" is taken.