All of lore.kernel.org
 help / color / mirror / Atom feed
* [LARTC] Latency in SFQ
@ 2001-12-18  7:07 Adi Nugroho
  2001-12-18  8:03 ` bert hubert
                   ` (3 more replies)
  0 siblings, 4 replies; 5+ messages in thread
From: Adi Nugroho @ 2001-12-18  7:07 UTC (permalink / raw)
  To: lartc

Dear all...

If we are using TBF, we can reduce "buffer" and "limit" to minimize latency.
What should we do in SFQ?

I tried to use SFQ, but the latency goes up to 10.000 to 20.000  milisecond 
in case of overloaded.

Can someone help me?

I'm using cbq.init, and here is my configuration:

Parent class:
DEVICE=eth0,10Mbit,1Mbit
RATEdKbit
WEIGHT=6.4Kbit
PRIO=5
LEAF=tbf
BUFFER000
LIMIT000
BOUNDED=yes
ISOLATED=no

In child class:
DEVICE=eth0,10Mbit,1Mbit
RATE\x16Kbit
WEIGHT=1.6Kbit
PRIO=5
PARENT\x0500
LEAF=sfq
BOUNDED=no
ISOLATED=no
PERTURB=5
QUANTUM000
RULE\x10.0.0.201

Regards,
Adi Nugroho

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

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [LARTC] Latency in SFQ
  2001-12-18  7:07 [LARTC] Latency in SFQ Adi Nugroho
@ 2001-12-18  8:03 ` bert hubert
  2001-12-18  8:45 ` Stef Coene
                   ` (2 subsequent siblings)
  3 siblings, 0 replies; 5+ messages in thread
From: bert hubert @ 2001-12-18  8:03 UTC (permalink / raw)
  To: lartc

On Tue, Dec 18, 2001 at 03:07:37PM +0800, Adi Nugroho wrote:
> Dear all...
> 
> If we are using TBF, we can reduce "buffer" and "limit" to minimize latency.
> What should we do in SFQ?

You can't really. SFQ has a fixed queue length of 128.

> I tried to use SFQ, but the latency goes up to 10.000 to 20.000  milisecond 
> in case of overloaded.

SFQ is no substitute for prioritization, but this is ridiculous. Can you
show the output of cbq.init? You can do this by looking into the script and
turning the logfile on.

I think cbq.init is misconfiguring things.

Regards,

bert


-- 
http://www.PowerDNS.com          Versatile DNS Software & Services
http://www.tk                              the dot in .tk
Netherlabs BV / Rent-a-Nerd.nl           - Nerd Available -
Linux Advanced Routing & Traffic Control: http://ds9a.nl/lartc

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

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [LARTC] Latency in SFQ
  2001-12-18  7:07 [LARTC] Latency in SFQ Adi Nugroho
  2001-12-18  8:03 ` bert hubert
@ 2001-12-18  8:45 ` Stef Coene
  2001-12-18 10:27 ` Adi Nugroho
  2001-12-18 10:34 ` bert hubert
  3 siblings, 0 replies; 5+ messages in thread
From: Stef Coene @ 2001-12-18  8:45 UTC (permalink / raw)
  To: lartc

> > I tried to use SFQ, but the latency goes up to 10.000 to 20.000 
> > milisecond in case of overloaded.
>
> SFQ is no substitute for prioritization, but this is ridiculous. Can you
> show the output of cbq.init? You can do this by looking into the script and
> turning the logfile on.
>
> I think cbq.init is misconfiguring things.
When I started with TC, I also tried the cbq.init script.  It's very limited 
in what you can do.  Once you know TC, you will realize that cbq.init is too 
limited and that's doing some things wrong because the situation is much more 
complicated.

An example.  You can select tbf as a leaf qfdisc for a cbq class.  But the 
qdisc is allready shaping, so attaching a tbf qdisc is not needed.  It can 
disturb the setup and can result in some strange behaviour.

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/

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [LARTC] Latency in SFQ
  2001-12-18  7:07 [LARTC] Latency in SFQ Adi Nugroho
  2001-12-18  8:03 ` bert hubert
  2001-12-18  8:45 ` Stef Coene
@ 2001-12-18 10:27 ` Adi Nugroho
  2001-12-18 10:34 ` bert hubert
  3 siblings, 0 replies; 5+ messages in thread
From: Adi Nugroho @ 2001-12-18 10:27 UTC (permalink / raw)
  To: lartc

On Tuesday 18 December 2001 16:03, bert hubert wrote:
>
> I think cbq.init is misconfiguring things.
>

Dear Sir,

As your suggestion, I have made all rule by hand.
Bellow are my script, one for cbq (tbf+sfq), and one for htb (+sfq)

But the latency was still so high (sometimes more than 20.000 ms if 
overloaded)

From statistic, the overlimit packet is high, but there is no dropped packet

Thank you for your help.

Regards,
Adi Nugroho

========Script for cbq (tbf+sfq)
========dev=eth1
DEVBWDT\x10Mbit
DEVWGHT=1Mbit
tc qdisc del dev eth1 root
tc qdisc add dev $dev root handle 1:0 cbq\
                bandwidth $DEVBWDT avpkt 1000 cell 8
tc class add dev $dev parent 1:0 classid 1:1 cbq \
                bandwidth $DEVBWDT rate $DEVBWDT weight $DEVWGHT \
                prio 8 allot 1514 cell 8 maxburst 20 avpkt 1000
 
#parent class
tc class add dev eth1 parent 1:1 classid 1:64 cbq \
        bandwidth 10Mbit rate 64Kbit weight 6.4Kbit prio 8\
        allot 1514 cell 8 maxburst 20 avpkt 1000 bounded
tc qdisc add dev eth1 parent 1:64 tbf \
                rate 64Kbit buffer 3000 limit 3000
 
#Child class #1
tc class add dev eth1 parent 1:64 classid 1:164 cbq \
        bandwidth 10Mbit rate 16Kbit weight 1.6Kbit prio 5 \
        allot 1514 cell 8 maxburst 20 avpkt 1000
 
tc qdisc add dev eth1 parent 1:164 sfq \
                perturb 5 quantum 4500
 
tc filter add dev eth1 parent 1:0 protocol ip \
                prio 100 u32 match ip dst 10.0.0.222 flowid 1:164
 
#Child class #2
tc class add dev eth1 parent 1:64 classid 1:264 cbq \
        bandwidth 10Mbit rate 16Kbit weight 1.6Kbit prio 5 \
        allot 1514 cell 8 maxburst 20 avpkt 1000
 
tc qdisc add dev eth1 parent 1:264 sfq \
                perturb 5 quantum 4500
 
tc filter add dev eth1 parent 1:0 protocol ip \
                prio 100 u32 match ip dst 10.0.0.73 flowid 1:264
 
#Child class #3
tc class add dev eth1 parent 1:64 classid 1:364 cbq \
        bandwidth 10Mbit rate 16Kbit weight 1.6Kbit prio 5 \
        allot 1514 cell 8 maxburst 20 avpkt 1000
 
tc qdisc add dev eth1 parent 1:364 sfq \
                perturb 5 quantum 4500
 
tc filter add dev eth1 parent 1:0 protocol ip \
                prio 100 u32 match ip dst 10.0.0.30 flowid 1:364


===========Script for htb (+sfq)
===========tc qdisc add dev eth1 root handle 1: htb
 
tc class add dev eth1 parent 1: classid 1:1 htb rate 64Kbit burst 15k
 
 
tc class add dev eth1 parent 1:1 classid 1:10 htb rate 16Kbit ceil 64Kbit 
burst 15k
tc class add dev eth1 parent 1:1 classid 1:20 htb rate 16Kbit ceil 64Kbit 
burst 15k
tc class add dev eth1 parent 1:1 classid 1:30 htb rate 16Kbit ceil 64Kbit 
burst 15k
 
 
tc qdisc add dev eth1 parent 1:10 handle 10: sfq perturb 10
tc qdisc add dev eth1 parent 1:20 handle 20: sfq perturb 10
tc qdisc add dev eth1 parent 1:30 handle 30: sfq perturb 10
 
u32="tc filter add dev eth1 protocol ip parent 1:0 prio 1 u32"
 
$u32 match ip dst 10.0.0.222 flowid 1:10
$u32 match ip dst 10.0.0.73 flowid 1:20
$u32 match ip dst 10.0.0.30 flowid 1:30

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

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [LARTC] Latency in SFQ
  2001-12-18  7:07 [LARTC] Latency in SFQ Adi Nugroho
                   ` (2 preceding siblings ...)
  2001-12-18 10:27 ` Adi Nugroho
@ 2001-12-18 10:34 ` bert hubert
  3 siblings, 0 replies; 5+ messages in thread
From: bert hubert @ 2001-12-18 10:34 UTC (permalink / raw)
  To: lartc

On Tue, Dec 18, 2001 at 06:27:51PM +0800, Adi Nugroho wrote:
> On Tuesday 18 December 2001 16:03, bert hubert wrote:
> >
> > I think cbq.init is misconfiguring things.
> >
> 
> Dear Sir,
> 
> As your suggestion, I have made all rule by hand.
> Bellow are my script, one for cbq (tbf+sfq), and one for htb (+sfq)
> 
> But the latency was still so high (sometimes more than 20.000 ms if 
> overloaded)

Lower maxburst for cbq and burst for htb.

> tc class add dev eth1 parent 1:1 classid 1:10 htb rate 16Kbit ceil 64Kbit 
> burst 15k

15k is ~8 seconds! 

Regards,

bert

-- 
http://www.PowerDNS.com          Versatile DNS Software & Services
http://www.tk                              the dot in .tk
Netherlabs BV / Rent-a-Nerd.nl           - Nerd Available -
Linux Advanced Routing & Traffic Control: http://ds9a.nl/lartc

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

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2001-12-18 10:34 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2001-12-18  7:07 [LARTC] Latency in SFQ Adi Nugroho
2001-12-18  8:03 ` bert hubert
2001-12-18  8:45 ` Stef Coene
2001-12-18 10:27 ` Adi Nugroho
2001-12-18 10:34 ` bert hubert

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.