| 摘要 |
A known congestion avoidance system for computer networks detects congestion at a node output port if the average queue length (integral) over the last congestion cycle plus the current (incomplete) cycle exceeds a fixed constant (taken as 1). (A congestion cycle is a period for which the queue length is 1 or more plus the following period for which the queue length is 0.) The time of arrival or departure of a message is stored at 21, the interval from the previous event is calculated at 22 and 23, the length of the current cycle is incremented at 25 by adding in the interval just determined, and the queue length at 26 is incremented or decremented by 1. The running integral for the current cycle is updated by having added into it the product formed at 27 of the interval since the last event (stored at 23) and the current queue length. The integrals for the current and previous cycles (stored at 24 and 30) are added and the lengths of those two cycles (stored at 29 and 31) are added, and the first sum divided at 34 by the second to obtain a grand average queue length. If that exceeds a preset value, then a congestion bit is set in messages leaving that node output port. In the present system, the running queue length average (in 29 min ) is maintained by adding (at 28 min ) the queue length (in 26 min ) into the average at regular intervals determined by timer ticks (from 60) (thus using integer addition instead of integer multiplication), and the grand average compared with the preset value by comparing (at 61) the total of the queue length averages with the total of the cycle periods (thus using integer addition and comparison instead of floating point operation). |
