Congestion control for high speed packet networks

Abstract

Abstract 5 n 7 Congestion in a high speed connection oriented packet network carrying bursty data and real time service is avoided by a novel in-call negotiation scheme for reserving bandwidth for real time calls and reserving buffer capacity for bursty calls. For a real time call, a centralized call controller in each switch in the network determines which of its outgoing trunks has available to it the peak bandwidth needed to accommodate the call. Once one of these trunks is found, the call routing is set up so that the call is directed through that trunk and the peak bandwidth requirement of the call is allocated on that output trunk for the duration of the call. Calls involving bursty data cause the call controllers in the switches to first set up the routing of the call and then allocate some minimum bandwidth on selected output trunks. A special packet then is produced by the call transmitter and sent to all the switches along the route which has just been set up. The special packet contains an indication of a requested buffer capacity. At each switch along the route between the origin and destination, a check is made to see if there is sufficient unallocated buffer capacity associated trunks along the route equal to the requested capacity. If there is sufficient unallocated buffer capacity, that amount of buffer capacity is allocated to the call and the special packet is passed unchanged to the next switch. If that amount of capacity is not available, then the switch may refuse the call by placing an indication of such refusal in the special packet and then it passes the packet to the next switch. The switch may also accept the call, if it has some minimal amount of unallocated capacity. In this case, it overwrites a lower amount of available buffer capacity into the special packet and sends it on to the next switch. The next switch, in turn, sees if it has an amount of unallocated buffer capacity equal to or greater than the requested buffer capacity now written into the special packet. The next switch behaves in the same way that previous switches behaved, either allocating the requested buffer capacity and passing the packet to the next switch unchanged, allocating some smaller amount of buffer capacity and overwriting the buffer capacity request in the special packet before passing it on to the next switch, or refusing the request because of insufficient capacity and placing an indication of such refusal in the special packet before passing it onto the next switch. When the special packet has reached the intended destination of the call, an acknowledgement packet is produced and sent back to the origin. If the request for buffer capacity has been completely denied by one of the switches or if a smaller amount of requested buffer capacity has been written into the special packet by one of the switches, the acknowledgement packet deallocates of the excess previously allocated buffer resources on its trip back to the origin. It also notifies the transmitter of the buffer capacity which has been allocated along the route of the call, if any. If the call has not been completely denied, the data making up the call is then sent along the prescribed route at a rate which will not cause an overflow of the buffers allocated to the call. The performance of the routing function is carried out in a centralized call controller in each switch and the performance of the reservation function is carried out in separate trunk controllers, each associated with one of the trunks connected to the switch. This unburdens the call controller of an extra function and permits the buffer reservation function to be accomplished in a shorter period of time than if it were to be performed by the call controller.

Inventors

• 👤 Bharat Tarachand Doshi
• 👤 Subrahmanyam Dravida
• 👤 David S. Einstein
• 👤 P. Harshavardhana

Bibliographic Details