Congestion Control
Generally, there are two types of congeston control. End-to-end congestion control and network-assisted congestion control (NACC). The formal does not require participation of the network for congestion control while the latter uses feedback infromation from the newtork for rate adjustment. In both types of congestion control, sender rate adjustment deals with the root of the problem.
For end-to-end congestion control, the feedback in terms of ACKs, the lack of feedback and the change in RTT can be used to detect congestion. The ideas are interesting and keep the underlying network simple. In constrast, network assisted congestion control requires the network to inform the sender the state of the network. The nodes may indicate the current state by marking the congestion indication bit, no increase bit or the explicit rate of a passing packet. While this additional information will obviously be useful, the computational complexity each node is higher. Because nodes have to maintain connection states, the scalability issue has to be addressed.
In NACC, the network takes on an active role in congestion control. For example, a node is fully aware of the current load and other states at its link, e.g. queue length, bandwidth, current utilization. With the limit bandwidth at an output link, the node can easily compute the fair bandwidth for flows sharing its link and inform all senders to adhere to the feedback of explicit rate. In a simplistic case, fair queueing is achieved. However, flows enter and leave the network according to user data behavior. The feedback could then be obselete information and the traffic at the link may start to oscillate. While the basic idea of this network assisted information may be correct, much work is needed to solve the problem of traffic dynamics. Other types feedback information could take the form of a congestion indication bit or a no increase bit. Is this complex problem worthy for researchers to ponder?
February 20th, 2010 at 5:56 pm
I want to thank the blogger very much not only for this post but also for his all previous efforts. I found http://www.innoqos.com to be greatly interesting. I will be coming back to http://www.innoqos.com for more information.