On Thu, 7 Jun 2001, Daniel Lee wrote:
>
> Message: 12
> From: Gery Kahn <geryk@sphera.com>
> To: "''lartc@mailman.ds9a.nl''"
<lartc@mailman.ds9a.nl>
> Date: Tue, 5 Jun 2001 20:28:30 +0200
> Subject: [LARTC] priority of class
>
> ->split traffic in 2 more classes
> ->tc class add dev eth0 parent 2:0 classid 2:21 cbq avpkt 1000 bandwidth
> ->100Mbit \
> ->rate 70Mbit prio 3 maxburst 20
>
> ->tc qdisc add dev eth0 parent 2:21 tbf rate 70Mbit buffer 45Kb/8 limit
15Kb
>
> ->tc class add dev eth0 parent 2:0 classid 2:22 cbq avpkt 1000 bandwidth
> ->100Mbit \
> ->rate 70Mbit prio 4 maxburst 20
>
> ->tc qdisc add dev eth0 parent 2:22 tbf rate 70Mbit buffer 45Kb/8 limit
15Kb
>
> ->Actualy results is Ok.I am trying to see influence of "prio"
parameter of
> ->class
> ->and nothing happened.
> ->Please, help me w/ your answer.
>
> You get the same conclusion as mine. I can''t see the effect of
class prio.
I don''t see why the prio parameter should work in this (incomplete?)
setup. I see 3 class levels. The prio parameters works if children of one
class have different prio values and the sum of the rates of the children
does not equal the parent rate.
Example:
class 1:1 rate 100Mbit
|
|-- class 1:11 rate 20Mbit prio 1
|
|-- class 1:12 rate 20Mbit prio 2
You will see that both child classes (1:11 and 1:12) will get a minimum of
20Mbit, but if both want more, class 1:11 will received the remaining
bandwidth first. So, in a high-load situation, class 1:11 gets 20 + 60Mbit
and class 1:12 will get only the minimum, 20Mbit.
Clear? ;-p