[j-nsp] CoS buffer size

[Dan Peachey]

On 24 June 2015 at 19:09, Saku Ytti <saku at ytti.fi> wrote:

> On (2015-06-24 16:08 +0100), Dan Peachey wrote:
>
> Hey Dan,
>
> > class-of-service {
> >     traffic-control-profiles {
> >         10M {
> >             scheduler-map 10M_COS;
> >             shaping-rate 10m;
>               guaranteed-rate 10m; # add this
> >         }
>

If I do this, can I still oversubscribe the interface? For example, lets
say I have 20 x IFL in an interface-set which I have shaped to 1G and each
of those 20 IFL's have 100M shaper with child CoS policy - so 2:1
oversubscription. My understanding was that I would need to use PIR/shaper
only for this and not touch G-rate since I cannot really offer a
"guaranteed rate" because of the oversubscription.


>
> > Queue    State        Max       Guaranteed   Burst  Weight Priorities
> > Drop-Rules
> > Index                 rate         rate      size            G    E
>  Wred
> >  Tail
> > ------ ----------- ----------- ------------ ------- ------ ----------
> > ----------
> >     32  Configured    10000000            0  131072    320   GL   EL    4
> >   0
> >     33  Configured    10000000            0  131072    320   GL   EL    4
> >   0
> >     34  Configured    10000000            0  131072     26   GL   EL    4
> >   0
> >     35  Configured    10000000            0  131072     13   GM   EH    4
> > 127
> >     36  Configured    10000000            0  131072      1   GL   EL    0
> > 255
> >     37  Configured    10000000      2400000  131072    320   GH   EH    4
> > 193
> >     38  Configured    10000000            0  131072      1   GL   EL    0
> > 255
> >     39  Configured    10000000            0  131072      1   GL   EL    0
> > 255
>
> Notice how g-rate is 0 for all but the policed class, all your queues
> running
> in excess area. Setting guarantee-rate == shaping-rate.
> Right now the % does not mean anything, only thing that means is what is
> excess priority (EL => excess low, EH => excess high, all EH traffic would
> be
> drained before any EL traffic is touched).
>
>
I thought the weights were determined by the %? The weights are then used
to schedule the queues appropriately. Even if the queues are in excess,
they should be weighted correctly?

Thanks,

Dan