Browse Prior Art Database

Flow-Balancing Congestion Control for Computer Networks

IP.com Disclosure Number: IPCOM000049500D
Original Publication Date: 1982-Jun-01
Included in the Prior Art Database: 2005-Feb-09
Document File: 4 page(s) / 65K

Publishing Venue

IBM

Related People

Anderson, H: AUTHOR [+4]

Abstract

A novel scheme for controlling buffer occupancy at nodes of a communication subnetwork is set forth. The scheme is suitable for networks with virtual circuit type of architecture; user conversations are assigned to a logical end-to-end full duplex path known as virtual circuit (VC). The VC is assumed to use a fixed route in the network during the course of a conversation and is flow-controlled by a window mechanism.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 40% of the total text.

Page 1 of 4

Flow-Balancing Congestion Control for Computer Networks

A novel scheme for controlling buffer occupancy at nodes of a communication subnetwork is set forth. The scheme is suitable for networks with virtual circuit type of architecture; user conversations are assigned to a logical end-to-end full duplex path known as virtual circuit (VC). The VC is assumed to use a fixed route in the network during the course of a conversation and is flow- controlled by a window mechanism.

The window flow control is implemented as follows: Consider a source which transmits messages to a destination. Upon receipt of a Go Ahead Response signal (GAHDRSP) which is an authorization from the destination, the source is allowed to send a window of w messages. Generally, a request for permission to send w messages is made by the source in the first message of every window. This request is called Go Ahead Request (GAHDRQ). The destination node responds to GAHDRQ by GAHDRSP based on resource availability at that node. In particular, a GAHDRSP may be sent to the source before the entire window of w messages has arrived. The window mechanism, therefore, limits the number of messages over a VC to 2w-1. SNA (System Network Architecture) uses this form of window mechanism. The number of messages a source is allowed to send at any instant will be denoted by PRCNT (Permit Count).

The basic idea behind flow-balancing congestion control is to control the traffic on each and every VC in the network, such that at each outgoing channel of nodes of the network the rate of in-flow in the VCs using this outgoing channel becomes equal to that of the outgoing channels at a node; the mechanism can then be applied to all outgoing channels at every node.

Consider the communication node A in Fig. 1 along with an outgoing channel. The communication node is on the path of several VCs which are all multiplexed over the outgoing channel. Messages which find the channel busy are temporarily stored in a queue, referred to as the out-queue; the queueing discipline is FCFS (first come, first serve). All virtual circuits are individually and independently flow-controlled by a window mechanism similar to the one described above. Consider one of these VCs and denote its window size by w. As described earlier, for every w message from the source of this VC a GAHDRQ is generated. The GAHDRSP for this VC transmitted from its destination passes through the communication node A.

In order to control the traffic, node A should constantly monitor the traffic on the incoming VCs. The means for doing this is by intercepting the GAHDRSPs returning to the sources, holding them in a hold queue (HLDQ), and releasing them upon certain conditions and availability of credits, as described below; this serves to throttle VCs. The mechanism is based on the following concepts: THRESHOLD-HOLD: In the out-queue, a threshold is specified;

we refer to this threshold by THLD (threshold-hold). When the

number of mess...