| Towards a Self Regulating Internet - SRI |
|
This is the official homepage of the SRI project. A project
exploring what control theory has to offer network communication systems.
The SRI project is supported by Vetenskapsrådet and Scint. |
|
A continually increasing use of the Internet with its widening set of services increases the demand on the control of the network. In industry, feedback control is a widely recognized method to increase productivity and product quality. Our intention is to explore what control theory has to offer network communication systems. The objective is to improve traffic throughput and to better accomodate different service demands. It has been experimentally observed that Internet traffic exhibits turbulent fluctuations. This causes congestion that results in delays and losses. Our goal is to develop control mechanisms for routers in heterogeneous networks to reduce these fluctuations. The approach is to design control strategies that operate with local information and yet optimize the system power (throughput divided by delay). We find support for this idea in the seminal works by Kelly [1 , 2]. More explictly, we propose to actively use the buffers in the routers to smoothen the traffic. This temporal control will be combined with spatial control, where the spatial distribution of the outgoing flow from a router is regulated using congestion information from other parts of the network. A fundamental issue in this context is to quantitatively relate the amount and spatial distribution of feedback information to the performance benefits that this information can bring. The main goal of this project is to develop control mechanisms in order to reduce traffic fluctuations. The considered control actuation is the buffers of the network. An efficient network would optimize the use of the buffers. Here we study such optimization both in the temporal and in the spatial domains. This will lead to two seperate control problems. We will also ask the natural question of what the system performance can gain from extra feedback information sent through the network. The project is consequently composed of three thrusts: temporal control, spatial control, and feedback complexity. |
|
Publications - The Team - Misc - Links |
|