Samsung Ubigate iBG2016, Руководство по настройке

Страница: 240 / 744

CHAPTER 10. QoS
204
© SAMSUNG Electronics Co., Ltd.
Congestion Management on Output Queues
In the iBG2016 system, RED(Random Early Detect) will be used to do
congestion avoidance on output queues. RED is the de facto standard for
congestion management in the internet today. The basic philosophy of RED is
to detect the onset of congestion and start dropping packets in a random
fashion before queue overflow leads to tail drops. Random drops not only
improves throughput of adaptive applications using TCP but is also more
suitable than tail drops for voice and streaming audio/video. Random drops
result in less perceptible degradation of voice/audio/video quality than
consecutive drops.
The intent of RED is detect persistent congestion but allow transient or
momentary bursts. In that sense, it behaves like a low pass filter.
RED therefore uses the ‘average queue’ size and not the ‘current queue size’
to determine congestion. The average queue size is calculated as an EWMA
(Exponential Weighted Moving Average) of the current queue size.
Simply put, the average queue size is a ‘smoothened-out’ version of the
instantaneous queue size.
The average queue size is compared against the ‘minimum threshold’ or
‘minth’ and ‘maximum threshold’ or ‘maxth’. When the average queue size is
less than minth, all packets are queued. When it is between minth and maxth,
packets are dropped based on the calculated drop probability. This drop
probability increases in direct proportion with the average queue size.
When the average queue size equals or exceeds maxth, all packets are dropped.
The intention here is to maintain the average queue size between minth and maxth.
The average queue size is calculated using the formula:
Qaverage = (1 - w) * Qa w * Qcurrent
Where, w is the weighting factor. It is a very small quantity and is configured
as a negative power of 2 for efficiency in calculation.
The drop probability when the average queue size equals ‘maxth’ is known as
the Mark Probability. Mark Probability = 1/MPD, where MPD is the Mark
Probability Denominator. MPD is configured as a power of 2 for efficiency in
Average Queue Size Calculation