Re: [Cake] How to test Cake on TP-Link WDR3600

[Alan Jenkins <alan.christopher.jenkins@gmail.com>]

On 26/07/15 22:20, Jonathan Morton wrote:
>
> Three reasons why a stackable peeler doesn't work so well:
>
> - There is some overhead from stacking, due to passing packets up and 
> down the stack.
>

> It also incurs at least one packet of unmanaged buffer.
>

> - A separate qdisc would not have the information about link bandwidth 
> and active flow count that cake maintains, and uses to influence the 
> peeler.
>

Sold!  Not that I'm a netdev reviewer, but it's a nice explanation of 
the design.

I.e. cake knows the minimum processing necessary to achieve the goal.  
The rest can still be left to hardware.

And now I recall peeling came from tbf, I guess it's not that odd.

Thanks for the illumination.
Alan

> - Peeling to individual IP packets is strictly necessary when 
> encapsulation overhead needs to be calculated.
>
> - Jonathan Morton
>