[LARTC] sum of child rates exceeds parent rate

[LARTC] sum of child rates exceeds parent rate

[jayesh rathod]

Hi,


i have created a parent class with 45Meg rate/ceiling


Note : The actual traffic flowing via that pc is around 6Meg to 15Meg
I assume the problem will come only when the traffic is more than 45Meg.


Now if i create child classes whose sum of rates crosses more than 45Meg ?
1.How that tc behaves.
2.Will this affect browsing.


Pls let me know this details

Regards
Jayesh
-------------------------------------------------
Still single? Click here to find the perfect match.

http://www.bharatmatrimony.com/cgi-bin/bmclicks1.cgi?141
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] sum of child rates exceeds parent rate

[Stef Coene]

On Monday 12 January 2004 06:40, jayesh rathod wrote:
> Hi,
>
> i have created a parent class with 45Meg rate/ceiling
>
> Note : The actual traffic flowing via that pc is around 6Meg to 15Meg
> I assume the problem will come only when the traffic is more than 45Meg.
Indeed.

> Now if i create child classes whose sum of rates crosses more than 45Meg ?
> 1.How that tc behaves.
It will try to give each class it's configured rate.  An other problem is the 
bottleneck.  YOU have to be the bottleneck and if you send more then your 
modem can handle, the modem will be the bottleneck and undo the traffic 
shaping.

> 2.Will this affect browsing.
It depends on the setup.  But yes, it can.  It's more likely it will affect 
real time traffic like icmp, VoIP, gaming, ...

Stef

-- 
stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.openprojects.net

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

RE: [LARTC] sum of child rates exceeds parent rate

[arek]

> > Note : The actual traffic flowing via that pc is around 6Meg to 15Meg
> > I assume the problem will come only when the traffic is more
> than 45Meg.
> Indeed.
>
> > Now if i create child classes whose sum of rates crosses more
> than 45Meg ?
> > 1.How that tc behaves.
> It will try to give each class it's configured rate.  An other
> problem is the
> bottleneck.  YOU have to be the bottleneck and if you send more
> then your
> modem can handle, the modem will be the bottleneck and undo the traffic
> shaping.

Wow wow, wait !

you can have 100 child classess in a sum of 100Megs, root class equal
10Megs.

the sum of all child classes will be 10Megs, and no more (if you ceil root
rate to 10Megs it at htb)

The behave of which child class get more /equal tokens than other you set
by priority parameter.

That is my theory with HTB+linux. With cbq many times total rate exceeds,
so i use it no more (it was not accurate). But HTB is accurate.

A.Binder

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] sum of child rates exceeds parent rate

[Stef Coene]

On Tuesday 13 January 2004 11:13, arek@chelmnet.pl wrote:
> > It will try to give each class it's configured rate.  An other
> > problem is the
> > bottleneck.  YOU have to be the bottleneck and if you send more
> > then your
> > modem can handle, the modem will be the bottleneck and undo the traffic
> > shaping.
>
> Wow wow, wait !
Ok :)

> you can have 100 child classess in a sum of 100Megs, root class equal
> 10Megs.
>
> the sum of all child classes will be 10Megs, and no more (if you ceil root
> rate to 10Megs it at htb)
Wrong.  The configured rate of a class is _always_ satisfied.  If you have a 
100M link, a parent class ceiled to 10M and 100 classes with rate = 1M, each 
class will get 1M.  So together they will get 100M.  And even if that is more 
the the ceil of the parent.
So you can overlimit a parent class.

> The behave of which child class get more /equal tokens than other you set
> by priority parameter.
Yes and no.  Each class will get his configured rate as a minimum.  If the 
parent class has some bandwidth left, it will be given to the class with the 
lowest prio.  
At the same time, the class with the lowest prio will be able to send it's 
packets first and so will have a lower delay.  BUT, if a low prio class uses 
more bandwidth then the configured rate, the latency goes up.

> That is my theory with HTB+linux. With cbq many times total rate exceeds,
> so i use it no more (it was not accurate). But HTB is accurate.
Yes, but trust me, you need to follow some rules.  You can find them on the 
faq page on docum.org.

Stef

-- 
stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.openprojects.net

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

RE: [LARTC] sum of child rates exceeds parent rate

[arek]

> > Wow wow, wait !
> Ok :)
> > you can have 100 child classess in a sum of 100Megs, root class equal
> > 10Megs.
> > the sum of all child classes will be 10Megs, and no more (if
> you ceil root
> > rate to 10Megs it at htb)
> Wrong.  The configured rate of a class is _always_ satisfied.
> If you have a
> 100M link, a parent class ceiled to 10M and 100 classes with
> rate = 1M, each
> class will get 1M.  So together they will get 100M.  And even if
> that is more
> the the ceil of the parent.
> So you can overlimit a parent class.

Well, i must practice that.
I've always thougght that root/parent queue tell lower queues to start
dropping packets.

Sure, you must be right, the queues will be told to drop packets, but they
will not do it unless they get their typed rate.
So if any of my 100 queues have 1Mbit traffic, then lower queues will start
to drop anything that is above 1Mbit for each queue individually.

So we overlimit 10Mbit celi about 10 times (in special case).


A.Binder

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] sum of child rates exceeds parent rate

[Stef Coene]

On Tuesday 13 January 2004 23:58, arek@chelmnet.pl wrote:
> > > Wow wow, wait !
> >
> > Ok :)
> >
> > > you can have 100 child classess in a sum of 100Megs, root class equal
> > > 10Megs.
> > > the sum of all child classes will be 10Megs, and no more (if
> >
> > you ceil root
> >
> > > rate to 10Megs it at htb)
> >
> > Wrong.  The configured rate of a class is _always_ satisfied.
> > If you have a
> > 100M link, a parent class ceiled to 10M and 100 classes with
> > rate = 1M, each
> > class will get 1M.  So together they will get 100M.  And even if
> > that is more
> > the the ceil of the parent.
> > So you can overlimit a parent class.
>
> Well, i must practice that.
> I've always thougght that root/parent queue tell lower queues to start
> dropping packets.
It's the other way around.  The class needs a token to send a packet.  As long 
as the class has tokens, it can send packets.  If the class has used all his 
tokens, it asks the parent if he has tokens left.

> Sure, you must be right, the queues will be told to drop packets, but they
> will not do it unless they get their typed rate.
Think about a bucket with tokens, not rate:
bucket size = burst
rate of new token entering bucket = rate
1 token = 1 packet
(this is for rate and ceil)

> So if any of my 100 queues have 1Mbit traffic, then lower queues will start
> to drop anything that is above 1Mbit for each queue individually.
Yes.

> So we overlimit 10Mbit celi about 10 times (in special case).
Yes.

Stef

-- 
stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.openprojects.net

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] sum of child rates exceeds parent rate

[Damjan]

> It's the other way around.  The class needs a token to send a packet.  As long 
> as the class has tokens, it can send packets.  If the class has used all his 
> tokens, it asks the parent if he has tokens left.

Hmm, then you should correct this:

http://www.docum.org/stef.coene/qos/tests/
    """
    If a child is using a token to send a packet, the same tocken is
    requested from the parent. So the child class is using the
    tokens/ctokens of it's parent. And without tokens, the parent can't give
    remaining bandwidth to it's child classes.
    """
And by the way, the next paragraph after that one is incomplete:
"...there is less traffic. That bur"


-- 
Damjan Georgievski
jabberID: damjan@bagra.net.mk
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] sum of child rates exceeds parent rate

[Damjan]

> > It's the other way around.  The class needs a token to send a packet.  As long 
> > as the class has tokens, it can send packets.  If the class has used all his 
> > tokens, it asks the parent if he has tokens left.
> 
> Hmm, then you should correct this:
> 
> http://www.docum.org/stef.coene/qos/tests/

Correction of the correction, the real URL is:
    http://www.docum.org/stef.coene/qos/tests/htb/burst/

PS.
one more example why frames in HTML are bad.


-- 
Damjan Georgievski
jabberID: damjan@bagra.net.mk
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] sum of child rates exceeds parent rate

[Stef Coene]

On Monday 19 January 2004 22:11, Damjan wrote:
> > It's the other way around.  The class needs a token to send a packet.  As
> > long as the class has tokens, it can send packets.  If the class has used
> > all his tokens, it asks the parent if he has tokens left.
>
> Hmm, then you should correct this:
>
> http://www.docum.org/stef.coene/qos/tests/
>     """
>     If a child is using a token to send a packet, the same tocken is
>     requested from the parent. So the child class is using the
>     tokens/ctokens of it's parent. And without tokens, the parent can't
> give remaining bandwidth to it's child classes.
>     """
It's not wrong.  The word "remaining" is important.  If the parent has no 
tokens left, it can't give remaining bandwidth.  But the child class can 
always send packets equal to the rate even if the parent has no tokens left 
so it can drag the parent (c)tokens negative.

Stef

-- 
stef.coene@docum.org
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.openprojects.net

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/