[c-nsp] Output Drops Due to QoS and threshold size

[Victor Sudakov]

Saku Ytti wrote:
> On 6 April 2016 at 20:09, Victor Sudakov <vas at mpeks.tomsk.su> wrote:
> 
> > Why is it *not* the default configuration?
> 
> I don't know why the defaults are so broken, small cats are worse with
> mls qos enabled than without. And the common wisdom is if you have
> microbursts MLS QoS will only make it worse, while counter-intuitively
> even with single queue with mls-qos you get more buffers than without
> it.

Still I don't understand the idea of configuring a threshold >100%. 

If for example IOS assigns 200 buffers to a port which are equally
distributed between 4 queues ("buffers 25 25 25 25"), whis means each
queue gets 50 buffers. If Thres1 in Q1 is 100%, it means 50 buffers. 

If Thres1 in Q1 is set to 1000%, this would mean 500 buffers. Where
are the 500-200=300 additional buffers borrowed? From the common pool?

-- 
Victor Sudakov,  VAS4-RIPE, VAS47-RIPN
sip:sudakov at sibptus.tomsk.ru