[RPRWG] Shaper credits withdrawal (down to loLimit)

["Stein Gjessing" <steing@xxxxxxxxx>]

In the process of running the simulation scenario given by David James and Harry Peng we have run into a question about shapers (referring to draft 2.3):

 

The text on page 109 (including figure 6.5) says that credits should be withdrawn (down to loLimit) when there are no frames waiting.

I believe this should be formalized in all the state tables for the shapers (tables 6.8 to 6.13)

 

In this context there seems to be a special problem with the duality of class A traffic: Is the A frame queued of class A0 or of class A1.   

 

This question is relevant two places:

 

It is relevant for the A0 and A1 shapers, but a guess the pragmatic answer is that the credits of these shapers are withdrawn (down to loLimit) only when there are no A frames waiting at all.

 

I believe this is a more serious question when it comes to the downstream shaper, shaperD:  In theory the credits of the D shaper should be withdrawn (down to loLimitD) when there are no C, B or A1 frames waiting. But how can the D shaper know if the A frames waiting are A0 frames?

 

One simple answer is that the withdrawal of D credits are irrespective of queued A frames?

Another simple answer is that the D credits are withdrawn when there are no frames waiting at all?

 

A much more complex answer is that the D shaper must check the value of the A0 and A1 shapers to see if the queued A frame is going to be sent as A0 or A1. But there might be more frames queued (some that will be sent as A0, some as A1).

 

Is there an answer to this in the draft already?

Comments?   Solutions  anybody?

 

Stein