[LARTC] quantum value for voip

[LARTC] quantum value for voip

[Victor Cassar]

Well

I understand why quantum should be at least equal to
mtu size

but let´s think in this

if i have a dsl link and i want to give more quality
to
"one voip user"  
i´ve already solved the dsl upload problem transfering
the bottleneck to my shapping box.

Should i change the quantum to a lower value for that
class?

any ideas or comments about this?

Thanks in advance



__________________________________________________
Do you Yahoo!?
Yahoo! Shopping - Send Flowers for Valentine's Day
http://shopping.yahoo.com
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] quantum value for voip

[Stef Coene]

On Thursday 13 February 2003 00:59, Victor Cassar wrote:
> Well
>
> I understand why quantum should be at least equal to
> mtu size
>
> but let´s think in this
>
> if i have a dsl link and i want to give more quality
> to
> "one voip user"
> i´ve already solved the dsl upload problem transfering
> the bottleneck to my shapping box.
>
> Should i change the quantum to a lower value for that
> class?
>
> any ideas or comments about this?
Do you know where quantum is used for?  It's for classes that are asking for 
remaining bandwidth.  So if that "one voip user" is never asking for more 
then the configured rate of its class, quantum is never used.  So it really 
depends on how you configured the classes to answer that question.

And taking a quantum of 10 byte is not "wrong".  But if you send 1000byte and 
your quantum is 10byte, htb will use 10byte in it's calculations.  So it can 
disturb the setup, but nothing bad will happen.

Stef

-- 

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

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

Re: [LARTC] quantum value for voip

[Victor Cassar]

Thanks Stef

Yep i missunderstood the quantum concept (it´s clear
now), but since my voip class was configured with
enough bw this doesn´t matter

Reading more on docum.org, i find hte possible cause
of the delays i was using sfq (128p queue) on the leaf
i removed sfq from the voip class


I will try also with other disciplines.. like tbf
just to have a deeper understanding of queuing in real
life applications

Are there any recomendation on which queuing
discipline to use with voip?
(my stats show 100 bytes per packet average in voip)

i´m doing more tests now and i´ll re re-read the full
lartc. doc

regards



--- Stef Coene <stef.coene@docum.org> wrote:
> On Thursday 13 February 2003 00:59, Victor Cassar
> wrote:
> > Well
> >
> > I understand why quantum should be at least equal
> to
> > mtu size
> >
> > but let´s think in this
> >
> > if i have a dsl link and i want to give more
> quality
> > to
> > "one voip user"
> > i´ve already solved the dsl upload problem
> transfering
> > the bottleneck to my shapping box.
> >
> > Should i change the quantum to a lower value for
> that
> > class?
> >
> > any ideas or comments about this?
> Do you know where quantum is used for?  It's for
> classes that are asking for 
> remaining bandwidth.  So if that "one voip user" is
> never asking for more 
> then the configured rate of its class, quantum is
> never used.  So it really 
> depends on how you configured the classes to answer
> that question.
> 
> And taking a quantum of 10 byte is not "wrong".  But
> if you send 1000byte and 
> your quantum is 10byte, htb will use 10byte in it's
> calculations.  So it can 
> disturb the setup, but nothing bad will happen.
> 
> Stef
> 
> -- 
> 
> stef.coene@docum.org
>  "Using Linux as bandwidth manager"
>      http://www.docum.org/
>      #lartc @ irc.oftc.net
> 
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO:
http://lartc.org/


__________________________________________________
Do you Yahoo!?
Yahoo! Shopping - Send Flowers for Valentine's Day
http://shopping.yahoo.com
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] quantum value for voip

[Stef Coene]

On Thursday 13 February 2003 17:00, Victor Cassar wrote:
> Thanks Stef
>
> Yep i missunderstood the quantum concept (it´s clear
> now), but since my voip class was configured with
> enough bw this doesn´t matter
>
> Reading more on docum.org, i find hte possible cause
> of the delays i was using sfq (128p queue) on the leaf
> i removed sfq from the voip class
>
> I will try also with other disciplines.. like tbf
> just to have a deeper understanding of queuing in real
> life applications
Tbf will not change nything.  Htb is acutally a tbf but with classes.

> Are there any recomendation on which queuing
> discipline to use with voip?
> (my stats show 100 bytes per packet average in voip)
Why not a simple fifo?  You can adapt the size of it from the command line.

> i´m doing more tests now and i´ll re re-read the full
> lartc. doc
Keep us informed.

Stef

-- 

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

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

Re: [LARTC] quantum value for voip

[Mathieu Deziel]

[-- Attachment #1: Type: text/plain, Size: 0 bytes --]



[-- Attachment #2: Type: text/html, Size: 1822 bytes --]

Re: [LARTC] quantum value for voip

[Mathieu Deziel]

Sorry, I sent it in HTML the first time.  Here it is again in text
format.

To minimize latency, (which is probably what you want for voip traffic),
give the class a high priority.
Make sure it never reaches its configured rate.  If it does, you will
see that the latency increases.
If you think it might reach its configured rate, than use a policing
filter before the class to place an upper limit on the rate.
And, as Steph memtionned, a FIFO is a good thing.  Just make sure it is
big enough to handle your expected burst, but not too big because it
could
increase the latency.

Good luck,
Mathieu.


  > Yep i missunderstood the quantum concept (it´s clear
  > now), but since my voip class was configured with
  > enough bw this doesn´t matter
  >
  > Reading more on docum.org, i find hte possible cause
  > of the delays i was using sfq (128p queue) on the leaf
  > i removed sfq from the voip class
  >
  > I will try also with other disciplines.. like tbf
  > just to have a deeper understanding of queuing in real
  > life applications
  Tbf will not change nything.  Htb is acutally a tbf but with classes.

  > Are there any recomendation on which queuing
  > discipline to use with voip?
  > (my stats show 100 bytes per packet average in voip)
  Why not a simple fifo?  You can adapt the size of it from the command
line.

  > i´m doing more tests now and i´ll re re-read the full
  > lartc. doc

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

Re: [LARTC] quantum value for voip

[Stef Coene]

On Friday 14 February 2003 01:18, Mathieu Deziel wrote:
> Sorry, I sent it in HTML the first time.  Here it is again in text
> format.
>
> To minimize latency, (which is probably what you want for voip traffic),
> give the class a high priority.
> Make sure it never reaches its configured rate.  If it does, you will
> see that the latency increases.
> If you think it might reach its configured rate, than use a policing
> filter before the class to place an upper limit on the rate.
I once suggested that as a solution, but has anyone tested it?

> And, as Steph 
:)

Stef

-- 

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

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

Re: [LARTC] quantum value for voip

[Mathieu Deziel]

>
> > If you think it might reach its configured rate, than use a policing
> > filter before the class to place an upper limit on the rate.
> I once suggested that as a solution, but has anyone tested it?
>

I did.  I works just great.  Very low latency.

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

[LARTC] quantum value for voip

[Victor Cassar]

1st thanks Stef and Mathieu 

wrote:
>Why not a simple fifo?  You can adapt the size of it
>from the command line.

I want tu use this pfifo queue i know its hardwired
and that i need to mark packets here or from the
source
in order to use dscp field

but i need some small sample to try

i cant figure out how to use this to divide traffic
in those 3 bands , or diferent rates

I´m analizing some sample scripts i took from the mail
archive, but i still need some hints, 

Do i need to use dsmark qdisc?

I greatly apreciate some samples from some one here
basicaly i need to divide traffic in 2 types voice &
and data, i already have a voip device with diffserv
capability, 

Another important thing is that i´m using
bridge+netfilter, so i´m not routing, 

regards & happy weekend
Victor



__________________________________________________
Do you Yahoo!?
Yahoo! Shopping - Send Flowers for Valentine's Day
http://shopping.yahoo.com
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] quantum value for voip

[Mathieu Deziel]

Hi Victor,  see below.


>
> >Why not a simple fifo?  You can adapt the size of it
> >from the command line.

> I want tu use this pfifo queue i know its hardwired
> and that i need to mark packets here or from the
> source
> in order to use dscp field
>

The pfifo and bfifo qdiscs are not "hardwired".  Their size is configurable (specified in bytes for the bfifo, and in number of packets for the pfifo).  The one that is hardwired is the pfifo_fast:  it is the default qdisc, the one that is there when you don't configure anything.  The pfifo_fast has 3 internal prio
queues.  Packets are classified in one of the three prio queue according to the dscp mark, just like you said in you previous mail.  Yes, to make use of the priority capability of pfifo_fast, your packets must marked before they enter the tc software.  You can use iptables to mark the packets before they enter tc.

Another thing you can do it make use of the prio qdisc, and then create your own (u32) filters.  You can then mark the packets with dsmark before the packets go out, if you want.

Hope this helps,
Mathieu.

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