[LARTC] CBQ inaccuracy

[LARTC] CBQ inaccuracy

[Amit Kucheria]

I have found CBQ to be very inaccurate (as others on the list)

I have created a structure as follow

  Root  (10mbit)
   |
  CBQ class (bandwidth=1mbit, rate\x10mbit)
   |
  CBQ qdisc (bandwidth=1mbit)

Now when i send excess data to the tune of 1.3mbit, cbq lets it thru.
If i increase it beyond that, CBQ suddenly wakes up and throttles it down 
to about 0.8Mbit!

Are there any explanations? If i am sending fixed size packets, should i 
set the 'avpkt' to that size?

I havent even able to use HTB correctly since a small burst size kills my 
thruput.

Any ideas about CBQ/HTB or anything which will throttle my bandwidth to 
1Mbit accurate *and* allow me to create qdiscs within. (Hence TBF is ruled 
out). My HTB script is given below i am no clue why it doesnt work.
 
-----------------------------------------------------------------------
BANDWIDTH="bandwidth 10Mbit"
LIMITBW="1Mbit"

$TC qdisc add $DEVICE root handle 1: htb

# This HTB class rate limits everyting to 1Mbit
$TC class add $DEVICE parent 1: classid 1:1 htb rate $LIMITBW ceil 
$LIMITBW burst 2k
----------------------------------------------------------------------
-- 
The statement below is true.
The statement above is false.
^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^
                  Amit Kucheria
          EECS Grad. Research Assistant
         University of Kansas @ Lawrence
   (R)+1-(785)-830 8521 ||| (O)+1-(785)-864 7774
____________________________________________________


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

Re: [LARTC] CBQ inaccuracy

[bert hubert]

On Wed, Dec 12, 2001 at 07:17:43PM -0600, Amit Kucheria wrote:
> I have found CBQ to be very inaccurate (as others on the list)
> 
> I have created a structure as follow
> 
>   Root  (10mbit)
>    |
>   CBQ class (bandwidth=1mbit, rate\x10mbit)
>    |
>   CBQ qdisc (bandwidth=1mbit)

Um, this doesn't make sense. I keep repeating this to posters here, SHOW
YOUR CONFIGURATION! Don't just draw pictures - your CBQ commandlines may not
do what you think they do.

In this case, it appears you've attached a qdisc within a qdisc or
something, I'm not sure. That could cause huge inaccuracies.

So show your configuration.

Regards,

bert

-- 
http://www.PowerDNS.com          Versatile DNS Software & Services
Trilab                                 The Technology People
Netherlabs BV / Rent-a-Nerd.nl           - Nerd Available -
'SYN! .. SYN|ACK! .. ACK!' - the mating call of the internet

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

Re: [LARTC] CBQ inaccuracy

[Amit Kucheria]

On Thu, 13 Dec 2001, bert hubert wrote:

> On Wed, Dec 12, 2001 at 07:17:43PM -0600, Amit Kucheria wrote:
> > I have found CBQ to be very inaccurate (as others on the list)
> > 
> > I have created a structure as follow
> > 
> >   Root  (10mbit)
> >    |
> >   CBQ class (bandwidth=1mbit, rate\x10mbit)
> >    |
> >   CBQ qdisc (bandwidth=1mbit)
> 
> Um, this doesn't make sense. I keep repeating this to posters here, SHOW
> YOUR CONFIGURATION! Don't just draw pictures - your CBQ commandlines may not
> do what you think they do.

My apologies. Here's the script

---------------------------------------------------------------------------
BANDWIDTHºndwidth 10Mbit
LIMITBW=1Mbit

# Root CBQ qdisc 1:
$TC qdisc add $DEVICE root handle 1: cbq $BANDWIDTH $AVPKT

# Root CBQ class 1:1
# This class rate limits everyting to 1Mbit
$TC class add $DEVICE parent 1:0 classid 1:1 cbq $BANDWIDTH rate $LIMITBW 
\
maxburst 100 $AVPKT allot 1514 weight 100Kbit prio 1 bounded isolated

# Filter packet from the 2 sources
$TC filter add $DEVICE parent 1: protocol ip prio 1 u32 match ip src 
$SRC1 flowid 1:1
$TC filter add $DEVICE parent 1: protocol ip prio 1 u32 match ip src 
$SRC2 flowid 1:1

# CBQ qdisc 2:0
$TC qdisc add $DEVICE parent 1:1 handle 2:0 cbq bandwidth $LIMITBW $AVPKT 
allot 1514
-------------------------------------------------------
-- 
The statement below is true.
The statement above is false.
^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^-^
                  Amit Kucheria
          EECS Grad. Research Assistant
         University of Kansas @ Lawrence
   (R)+1-(785)-830 8521 ||| (O)+1-(785)-864 7774
____________________________________________________


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

Re: [LARTC] CBQ inaccuracy

[Stef Coene]

On Thursday 13 December 2001 08:50, you wrote:
> On Thu, 13 Dec 2001, bert hubert wrote:
> > On Wed, Dec 12, 2001 at 07:17:43PM -0600, Amit Kucheria wrote:
> > > I have found CBQ to be very inaccurate (as others on the list)
> > >
> > > I have created a structure as follow
> > >
> > >   Root  (10mbit)
> > >
> > >   CBQ class (bandwidth=1mbit, rate=10mbit)
> > >
> > >   CBQ qdisc (bandwidth=1mbit)
> >
> > Um, this doesn't make sense. I keep repeating this to posters here, SHOW
> > YOUR CONFIGURATION! Don't just draw pictures - your CBQ commandlines may
> > not do what you think they do.
>
> My apologies. Here's the script
>
> ---------------------------------------------------------------------------
> BANDWIDTH=bandwidth 10Mbit
> LIMITBW=1Mbit
>
> # Root CBQ qdisc 1:
> $TC qdisc add $DEVICE root handle 1: cbq $BANDWIDTH $AVPKT
>
> # Root CBQ class 1:1
> # This class rate limits everyting to 1Mbit
> $TC class add $DEVICE parent 1:0 classid 1:1 cbq $BANDWIDTH rate $LIMITBW
> \
> maxburst 100 $AVPKT allot 1514 weight 100Kbit prio 1 bounded isolated
>
> # Filter packet from the 2 sources
> $TC filter add $DEVICE parent 1: protocol ip prio 1 u32 match ip src
> $SRC1 flowid 1:1
> $TC filter add $DEVICE parent 1: protocol ip prio 1 u32 match ip src
> $SRC2 flowid 1:1
>
> # CBQ qdisc 2:0
> $TC qdisc add $DEVICE parent 1:1 handle 2:0 cbq bandwidth $LIMITBW $AVPKT
> allot 1514
> -------------------------------------------------------
I'm not seeing any big errors.  I tried also the bounded parameter of CBQ, 
you can find the results on docum.org.  I was able to throttle the bandwidth 
at each speed I want.  I din't attach the second CBQ, but only used the 
bounded class, maybe you can try again without the second CBQ.  
And don't use the isolated parameter.  It's not working and sometimes it can 
disturb the configuration.


Stef

-- 

stef.coene@docum.org
 More QOS info : http://docum.org/
 Title : "Using Linux as bandwidth manager"
     

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

Re: [LARTC] CBQ inaccuracy

[bert hubert]

On Thu, Dec 13, 2001 at 09:20:32AM +0100, Stef Coene wrote:

> > > Um, this doesn't make sense. I keep repeating this to posters here, SHOW
> > > YOUR CONFIGURATION! Don't just draw pictures - your CBQ commandlines may
> > > not do what you think they do.
> >
> > My apologies. Here's the script

Amit, no apologies are needed, it's just that it helps to get good answers
:-)

> > # CBQ qdisc 2:0
> > $TC qdisc add $DEVICE parent 1:1 handle 2:0 cbq bandwidth $LIMITBW $AVPKT
> > allot 1514


Just don't do this. Having a CBQ within a CBQ is useless and can indeed
cause weird behaviour. You will get two qdiscs 'stacked' so to speak, both
trying to shape traffic. At times, the inner qdisc will not want to give up
data, which doesn't make life easier on the outer one.

CBQ can do shaping out of its own accord - it doesn't need a second shaper
to do that for it. For the same reasone, people should not embed TBFs in
CBQ!

Regards,

bert

-- 
http://www.PowerDNS.com          Versatile DNS Software & Services
Trilab                                 The Technology People
Netherlabs BV / Rent-a-Nerd.nl           - Nerd Available -
'SYN! .. SYN|ACK! .. ACK!' - the mating call of the internet

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