[LARTC] Complex Traffic shaping setup
George Alexandru Dragoi
George Alexandru Dragoi <waruiinu@gmail.com>
Fri Feb 18 13:31:51 CET 2005
I'm not sure if you have to use different prio for filters (With fw
this is a must). Depends on your setup.
With htb, everything is attached with parent as root qdisc.
On Fri, 18 Feb 2005 18:27:26 +0530, Padam J Singh
<padam.singh@inventum.cc> wrote:
> Hi!
>
>
> I have the following setup using iproute2:
>
> ========
> | Root |
> ========
> |
> | ===============
> ------------| Box Level 1 |
> | ===============
> |
> | ===============
> ------------| Box Level 2 |
> | ===============
> |
> |
> | ===============
> ------------| Users |
> ===============
> |
> | ===============
> ------------| User 1 |
> | ===============
> |
> | ===============
> ------------| User 2 |
> | ===============
> |
> | ===============
> ------------| User 3 |
> ===============
>
>
> I am using HTB for each node, simply because it supports burst mode.
>
> There are box level classes (top level classes) that might limit bandiwdth
> based on protocols (http, smtp etc)
> Then there are users classes which rate limit each user to a specific rate.
>
> Each Box Level and User Classes have associated Qdiscs as well.
>
> The issue is that HTB seems to be rate limiting only if I attach filters at
> the root (1:).
>
> Since classification for box level classes may also match more specific
> user filters are well, one of the two things happen:
>
> 1. If the prio of the filters for Box level is higher or equal than that of
> Users, packets are only sent to the box level class, and since the rate at
> box level class is generally higher, the user starts exceeding his limit.
> 2. If the prio of the User filter is higher, packets matching the user
> filter are never sent to the Box level classes, making the box level rate
> for a particular protocol exceed the required for all users put together.
>
> Is there a solutions to this issue? I am unable to really find a solution
> since filters can only be applied to the root!
>
> Thanks in advance!
> Padam Singh.
>
>
--
Bla bla
More information about the LARTC
mailing list