Effect of virtual end systems for rate-based congestion control

In data transmission on a broadband network, such as an ATM network, rate-b ased congestion control is considered effective in preventing data loss cau sed by buffer overflow at a relay node and in achieving high throughput. Un der this control, the sender node dynamically changes the transmission rate according to the network condition. However as the network grows, the dela y for inquiry to the network condition increases so that it becomes hard to control congestion adequately. One of the solutions to that problem is to divide the network into several logical blocks and then to install virtual end systems at the block boundaries. When congestion control is carried out at each block, the delay for inquiry is decreased and the queue length of each node is kept low. As a result, cell discard caused by buffer overflow is reduced. In this paper, we build a queuing network model and then analyz e it approximately to evaluate the effect of the virtual end systems. We in vestigate the influence of delay between the two ends and the locations of the virtual end systems in the network and then find some situations in whi ch the virtual end systems do not work efficiently. The result shows that i t is not sufficient simply to increase the number of virtual end systems an d/or to consider their locations. The virtual end system should report its rate reduction to the sender node and should reduce its rate reduction rati o to serve more cells in its buffer. (C) 1999 Scripta Technica.