From: Joe Chang (changjoe@earthlink.net)
Date: Sun May 23 2004 - 18:29:44 GMT-3
> Yes, once you start queuing, you keep queueing (you are not supposed to
> reorder traffic in traffic shapping) and then transmit from the queue to
> the tx-ring as much as Bc per Tc. Eventually you either empty the queue
> and fall back to initial strategy or you start tail-dropping excess
> traffic that does not fit your shapping buffer.
That makes sense. Until I come across a better explanation, this is the one
I'll stick to . Thanks.
> All shapping metrics are done based on Tc. Tc is calculated from rate
> and Bc. Then there is that shape to average or peak. I don't really know
> what the difference between shaping to average with be = 0 and shapping
> to peak would be though, but it comes handy to easily define different
> shapping schemes. (AFAIK, shaping to peak with given (Bc,Be) is the same
> as shapping to average with (Bc+Be,0).
If Be is 0 then the shaping is as aggressive as possible -- at no point
would the real rate be greater than Bc in a Tc period of time. If Be is set
such that the total bucket size is equal to the peak rate divided by Tc,
then you are allowing an occassional burst to go to maximum capacity of the
interface.
> One thing that seems not clear is that Be does not apply to a Tc, but
> instead is the size of your "overrun" quota. That quota is
> "rechargeable" so to speak, as long and you drop under cir.
In other words shaping is a "save now, spend more later" approach. Which is
differeent from traffic policing -- "spend more now, pay it back later".
This archive was generated by hypermail 2.1.4 : Wed Jun 02 2004 - 11:12:15 GMT-3