Samsung Ubigate iBG3026, Руководство по настройке

Страница: 236 / 718

CHAPTER 19. QoS
198
© SAMSUNG Electronics Co., Ltd.
Also the numerator in the above expression, which is fixed for a given
configuration,((maxTh - minTh)/ 2mpd), is pre-calculated at the time of
initialization/configuration. This saves CPU time in the forwarding path.
The above drop rate procedure is same for both interfaces, and traffic classes.
SWRED
DSWRED allows specifying the separate set of RED parameters(only minth
and maxth thresholds) for each DSCP value. The weight factor(ewf), and
mark probability denominator(mpd) is same for all DSCP values. But the
maxth and minth thresholds can be a different for each DSCP value.
As said earlier by default RED is enabled on traffic classes. At the queuing
time, if DSWRED is not enabled then minth and maxth thresholds of given
class is assumed for RED calculation otherwise the minth and maxth
thresholds corresponding to the packet’s DSCP value is assumed.
Configuring RED parameters
This section discusses which RED parameters are configurable and how are
their default values set. It also explains the effect of changing these parameter
values and guidelines about how to set their values.
interface RED parameters
The EWF parameter determines the weight factor in the calculation of the
average queue size. The weight factor ‘w’ is calculated using the formula(w =
1/(2^ewf)). The EWF parameter determines how RED is sensitive to traffic
fluctuations or bursts. If EWF is very low, then the RED is very sensitive to
traffic fluctuations and causes un-necessary early random packet drops and
there by reducing network throughput. If EWF is very high, then the RED is
not sensitive to traffic fluctuations and causes frequent tail drops which
reduces the effectiveness of RED. Therefore it is important to strike a good
balance in the value of EWF. The default value of ewf will be 5, which means
the actual EWF value = 1/32.
The MPD value determines the drop probability when the average queue size
is close to maxth. A high value will result in more aggressive drops for a given
increase in average queue size and could potentially lead to queue size
oscillations. A low value will result in a milder drop response for a given
increase in average queue size and increases the likelihood of tail drops.