Prev: Google-Earth = non-Open-source ?
Next: wifi routing problem
From: Andy Furniss on 4 May 2008 16:37
Rafa� M wrote:
> Hello,
>
> tc class add dev ifb0 parent 1:1 classid 1:2 hfsc sc m2 450kbit ul m2
> 512kbit
> tc class add dev ifb0 parent 1:1 classid 1:3 hfsc sc m2 62kbit ul m2
> 512kbit
>
> why i don't have guaranteed bandwidth 450 and 62? when i set ul m2 450
> and 62 then bandwidth is splited OK, but when i shutdown downloading
> 1:2, 1:3 not go to 512
>
> please help!

I never really used hfsc, I did some tests once and found it's not
really quite like the hfsc in the sigcom paper in that it relies on
individual rt classes being backlogged rather than the link, which means
you can't get it to be packet perfect for slow links, so I kind of gave
up trying to learn how to configure it, as it seemed like a waste of
time for shaping on a slow wan, though people who shape lan would
probably like it the way it is.

To totally contradict myself - depending on setup I think you could make
it behave better than htb if you have many bulk classes as long as you
use ls classes - which you should for bulk, not that any of the examples
I've seen seem to.

If you ever use it on eth rather than ifb remember it will by default
drop unclassified traffic - so either make a default class or remember
to tc filter arp.

As for your setup - try without the ul on the leafs but add sc rate
512kbit ul rate 512kbit to the parent class 1:1.

Andy.
From: Rafał M on 5 May 2008 11:44
On 4 Maj, 22:37, Andy Furniss <s...(a)andyfurniss.entadsl.com> wrote:
> Rafa³ M wrote:
> > Hello,
>
> > tc class add dev ifb0 parent 1:1 classid 1:2 hfsc sc m2 450kbit ul m2
> > 512kbit
> > tc class add dev ifb0 parent 1:1 classid 1:3 hfsc sc m2 62kbit ul m2
> > 512kbit
>
> > why i don't have guaranteed bandwidth 450 and 62? when i set ul m2 450
> > and 62 then bandwidth is splited OK, but when i shutdown downloading
> > 1:2, 1:3 not go to 512
>
> > please help!
>
> I never really used hfsc, I did some tests once and found it's not
> really quite like the hfsc in the sigcom paper in that it relies on
> individual rt classes being backlogged rather than the link, which means
> you can't get it to be packet perfect for slow links, so I kind of gave
> up trying to learn how to configure it, as it seemed like a waste of
> time for shaping on a slow wan, though people who shape lan would
> probably like it the way it is.
>
> To totally contradict myself - depending on setup I think you could make
> it behave better than htb if you have many bulk classes as long as you
> use ls classes - which you should for bulk, not that any of the examples
> I've seen seem to.
>
> If you ever use it on eth rather than ifb remember it will by default
> drop unclassified traffic - so either make a default class or remember
> to tc filter arp.
>
> As for your setup - try without the ul on the leafs but add sc rate
> 512kbit ul rate 512kbit to the parent class 1:1.
>
> Andy.

very thanks
 | 
Pages: 1
Prev: Google-Earth = non-Open-source ?
Next: wifi routing problem