[LARTC] htb2 -> htb3 problems

[LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

Hello!

I need to switch from htb2 to htb3, because of speed issues (for me,
htb2 is unable to handle more then 100mbit duplex with ~550 classes),
kernel profiling shows htb_dequeue_prio at 1st place with 3x isolation.

So, I've moved from 2.4.19 to 2.4.25 kernel (hi-pac for classification/marking
and htb3 for queueing), and traffic rate drop from 100 to 20mbit.

What can be wrong? The only change I see is htb2 -> htb3


Here is my qdiscs/classes/filters example

qdisc htb 1: dev eth0 r2q 10 default 2500 direct_packets_stat 4604 ver 3.13
qdisc sfq 100: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3300: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3301: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3302: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3304: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3305: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3306: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
qdisc sfq 3308: dev eth0 quantum 1514b limit 128p flows 128/1024 perturb 5sec
[skip]


class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
class htb 1:3300 parent 1:1 leaf 3300: prio 0 quantum 13107 rate 1Mbit ceil 1126Kbit burst 1023b/8 mpu 0b cburst 0b/8 mpu 0b level 0
class htb 1:3301 parent 1:1 leaf 3301: prio 0 quantum 10240 rate 800Kbit ceil 1Mbit burst 1023b/8 mpu 0b cburst 0b/8 mpu 0b level 0
class htb 1:3302 parent 1:1 leaf 3302: prio 0 quantum 13107 rate 1Mbit ceil 1200Kbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
class htb 1:3304 parent 1:1 leaf 3304: prio 0 quantum 4915 rate 384Kbit ceil 384Kbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
[skip]


filter parent 1: protocol ip pref 50 fw
filter parent 1: protocol ip pref 50 fw handle 0x14b4 classid 1:5300
filter parent 1: protocol ip pref 50 fw handle 0x14b5 classid 1:5301
filter parent 1: protocol ip pref 50 fw handle 0x14b6 classid 1:5302
filter parent 1: protocol ip pref 50 fw handle 0x14b8 classid 1:5304
filter parent 1: protocol ip pref 50 fw handle 0x14b9 classid 1:5305
filter parent 1: protocol ip pref 50 fw handle 0x14ba classid 1:5306
[skip]

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

Re: [LARTC] htb2 -> htb3 problems

[devik]

> I need to switch from htb2 to htb3, because of speed issues (for me,
> htb2 is unable to handle more then 100mbit duplex with ~550 classes),
> kernel profiling shows htb_dequeue_prio at 1st place with 3x isolation.
>
> So, I've moved from 2.4.19 to 2.4.25 kernel (hi-pac for classification/marking
> and htb3 for queueing), and traffic rate drop from 100 to 20mbit.
>
> What can be wrong? The only change I see is htb2 -> htb3

Hello,

I suppose the drop you see is CPU bound ? Did you profiled it
again ?
Both HTB algorithms are very different  so that one can expect
different behavior with different data/rules. But I can admit
that this 5x drop is rather big and unfortunate.

devik

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

Re: [LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

devik (devik@cdi.cz) wrote:
> > I need to switch from htb2 to htb3, because of speed issues (for me,
> > htb2 is unable to handle more then 100mbit duplex with ~550 classes),
> > kernel profiling shows htb_dequeue_prio at 1st place with 3x isolation.
> >
> > So, I've moved from 2.4.19 to 2.4.25 kernel (hi-pac for classification/marking
> > and htb3 for queueing), and traffic rate drop from 100 to 20mbit.
> >
> > What can be wrong? The only change I see is htb2 -> htb3
> 
> Hello,
> 
> I suppose the drop you see is CPU bound ? Did you profiled it
> again ?

No, CPU is about times more in idle state in htb3, and 
5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
reason of moving to htb3. 

> Both HTB algorithms are very different  so that one can expect
> different behavior with different data/rules. But I can admit
> that this 5x drop is rather big and unfortunate.

any ideas?

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

Re: [LARTC] htb2 -> htb3 problems

[devik]

> > I suppose the drop you see is CPU bound ? Did you profiled it
> > again ?
>
> No, CPU is about times more in idle state in htb3, and
> 5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
> reason of moving to htb3.

ok then it is config issue probably. you should pin it down
to smallest possible number of classes for test (say up to
5 classes) and then use tc -s show class ... to see internal
statistics. Look for classes with small (or negative)
tokens or ctokens - these are in throttling state and are
slowing throughtput - then think if it is ok ..

From data you send one can't decide what's bad. In any case
and as I said before, behavioue changed a bit so that results
can be a bit different a may need tc script changes.

devik

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

Re: [LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

devik (devik@cdi.cz) wrote:
> > > I suppose the drop you see is CPU bound ? Did you profiled it
> > > again ?
> >
> > No, CPU is about times more in idle state in htb3, and
> > 5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
> > reason of moving to htb3.
> 
> ok then it is config issue probably. you should pin it down
> to smallest possible number of classes for test (say up to
> 5 classes) and then use tc -s show class ... to see internal
> statistics. Look for classes with small (or negative)
> tokens or ctokens - these are in throttling state and are
> slowing throughtput - then think if it is ok ..

ok
I'm creating root with 200Mbit
and parent with 10Mbit/1Mbit


class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 131072 rate 10Mbit ceil 12Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
 Sent 25443954 bytes 17155 pkts (dropped 0, overlimits 0)
 rate 143050bps 97pps backlog 25p
 lended: 17130 borrowed: 0 giants: 0
 tokens: 335 ctokens: -787

class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
 Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
 rate 2723bps 49pps
 lended: 0 borrowed: 0 giants: 0
 tokens: 8241 ctokens: 8241

class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
 Sent 25406104 bytes 17130 pkts (dropped 0, overlimits 0)
 rate 144147bps 98pps
 lended: 0 borrowed: 0 giants: 0
 tokens: 8195 ctokens: 8195

class htb 1:3500 parent 1:1 leaf 3500: prio 0 quantum 13107 rate 1Mbit ceil 1Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
 Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
 rate 2739bps 49pps
 lended: 8712 borrowed: 0 giants: 0
 tokens: 12501 ctokens: -294


so, rate is 1,2Mbit
and what is meaning of negative ctokens?
 
> >From data you send one can't decide what's bad. In any case
> and as I said before, behavioue changed a bit so that results
> can be a bit different a may need tc script changes.
> 
> devik
> 
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

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

Re: [LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

Michael Vasilenko (acid@dg.net.ua) wrote:
> devik (devik@cdi.cz) wrote:
> > > > I suppose the drop you see is CPU bound ? Did you profiled it
> > > > again ?
> > >
> > > No, CPU is about times more in idle state in htb3, and
> > > 5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
> > > reason of moving to htb3.
> > 
> > ok then it is config issue probably. you should pin it down
> > to smallest possible number of classes for test (say up to
> > 5 classes) and then use tc -s show class ... to see internal
> > statistics. Look for classes with small (or negative)
> > tokens or ctokens - these are in throttling state and are
> > slowing throughtput - then think if it is ok ..
> 
> ok
> I'm creating root with 200Mbit
> and parent with 10Mbit/1Mbit

and qdisc stats (I have sfq attached to each leaf htb class):

 
qdisc sfq 5500: quantum 1514b limit 128p flows 128/1024 perturb 5sec
 Sent 52611108 bytes 35988 pkts (dropped 0, overlimits 0)
 backlog 22p

 qdisc sfq 3500: quantum 1514b limit 128p flows 128/1024 perturb 5sec
 Sent 1004864 bytes 18073 pkts (dropped 0, overlimits 0)

 qdisc htb 1: r2q 10 default 2500 direct_packets_stat 130 ver 3.13
 Sent 53627322 bytes 54191 pkts (dropped 0, overlimits 89908)
 backlog 22p

 
> class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 131072 rate 10Mbit ceil 12Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
>  Sent 25443954 bytes 17155 pkts (dropped 0, overlimits 0)
>  rate 143050bps 97pps backlog 25p
>  lended: 17130 borrowed: 0 giants: 0
>  tokens: 335 ctokens: -787
> 
> class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
>  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
>  rate 2723bps 49pps
>  lended: 0 borrowed: 0 giants: 0
>  tokens: 8241 ctokens: 8241
> 
> class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
>  Sent 25406104 bytes 17130 pkts (dropped 0, overlimits 0)
>  rate 144147bps 98pps
>  lended: 0 borrowed: 0 giants: 0
>  tokens: 8195 ctokens: 8195
> 
> class htb 1:3500 parent 1:1 leaf 3500: prio 0 quantum 13107 rate 1Mbit ceil 1Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
>  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
>  rate 2739bps 49pps
>  lended: 8712 borrowed: 0 giants: 0
>  tokens: 12501 ctokens: -294
> 
> 
> so, rate is 1,2Mbit
> and what is meaning of negative ctokens?
>  
> > >From data you send one can't decide what's bad. In any case
> > and as I said before, behavioue changed a bit so that results
> > can be a bit different a may need tc script changes.
> > 
> > devik
> > 
> > _______________________________________________
> > LARTC mailing list / LARTC@mailman.ds9a.nl
> > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
> 
> -- 
> Michael Vasilenko

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

Re: [LARTC] htb2 -> htb3 problems

[devik]

I see you have cburst 0 ! It is not allowed
and is described in docs. What was commands to
create the classes ?

-------------------------------
    Martin Devera aka devik
Linux kernel QoS/HTB maintainer
  http://luxik.cdi.cz/~devik/

On Tue, 6 Apr 2004, Michael Vasilenko wrote:

> devik (devik@cdi.cz) wrote:
> > > > I suppose the drop you see is CPU bound ? Did you profiled it
> > > > again ?
> > >
> > > No, CPU is about times more in idle state in htb3, and
> > > 5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
> > > reason of moving to htb3.
> >
> > ok then it is config issue probably. you should pin it down
> > to smallest possible number of classes for test (say up to
> > 5 classes) and then use tc -s show class ... to see internal
> > statistics. Look for classes with small (or negative)
> > tokens or ctokens - these are in throttling state and are
> > slowing throughtput - then think if it is ok ..
>
> ok
> I'm creating root with 200Mbit
> and parent with 10Mbit/1Mbit
>
>
> class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 131072 rate 10Mbit ceil 12Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
>  Sent 25443954 bytes 17155 pkts (dropped 0, overlimits 0)
>  rate 143050bps 97pps backlog 25p
>  lended: 17130 borrowed: 0 giants: 0
>  tokens: 335 ctokens: -787
>
> class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
>  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
>  rate 2723bps 49pps
>  lended: 0 borrowed: 0 giants: 0
>  tokens: 8241 ctokens: 8241
>
> class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
>  Sent 25406104 bytes 17130 pkts (dropped 0, overlimits 0)
>  rate 144147bps 98pps
>  lended: 0 borrowed: 0 giants: 0
>  tokens: 8195 ctokens: 8195
>
> class htb 1:3500 parent 1:1 leaf 3500: prio 0 quantum 13107 rate 1Mbit ceil 1Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
>  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
>  rate 2739bps 49pps
>  lended: 8712 borrowed: 0 giants: 0
>  tokens: 12501 ctokens: -294
>
>
> so, rate is 1,2Mbit
> and what is meaning of negative ctokens?
>
> > >From data you send one can't decide what's bad. In any case
> > and as I said before, behavioue changed a bit so that results
> > can be a bit different a may need tc script changes.
> >
> > devik
> >
> > _______________________________________________
> > LARTC mailing list / LARTC@mailman.ds9a.nl
> > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
>
> --
> Michael Vasilenko
>
>

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

Re: [LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

devik (devik@cdi.cz) wrote:
> I see you have cburst 0 ! It is not allowed
> and is described in docs. What was commands to
> create the classes ?

 
/sbin/tc.3 qdisc add dev eth0 root handle 1:0 htb default 2500 r2q 100
/sbin/tc.3 class add dev eth0 parent 1:0 classid 1:1 htb rate 200mbit ceil 200mbit quantum 200000
/sbin/tc.3 class add dev eth0 parent 1:0 classid 1:2 htb rate 200mbit ceil 200mbit quantum 200000
/sbin/tc.3 class add dev eth0 parent 1:1 classid 1:3500 htb rate 10Mbit ceil 10Mbit burst 2048b cburst 1
/sbin/tc.3 qdisc add dev eth0 parent 1:3500 handle 3500: sfq perturb 5
/sbin/tc.3 class add dev eth0 parent 1:2 classid 1:5500 htb rate 20Mbit ceil 20Mbit burst 2048b cburst 1 quantum 60000
/sbin/tc.3 qdisc add dev eth0 parent 1:5500 handle 5500: sfq perturb 5
/sbin/tc.3 filter add dev eth0 parent 1:0 protocol ip prio 50 handle 3500 fw classid 1:3500
/sbin/tc.3 filter add dev eth0 parent 1:0 protocol ip prio 50 handle 5500 fw classid 1:5500



tc.3 -V
tc utility, iproute2-ss020116

> On Tue, 6 Apr 2004, Michael Vasilenko wrote:
> 
> > devik (devik@cdi.cz) wrote:
> > > > > I suppose the drop you see is CPU bound ? Did you profiled it
> > > > > again ?
> > > >
> > > > No, CPU is about times more in idle state in htb3, and
> > > > 5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
> > > > reason of moving to htb3.
> > >
> > > ok then it is config issue probably. you should pin it down
> > > to smallest possible number of classes for test (say up to
> > > 5 classes) and then use tc -s show class ... to see internal
> > > statistics. Look for classes with small (or negative)
> > > tokens or ctokens - these are in throttling state and are
> > > slowing throughtput - then think if it is ok ..
> >
> > ok
> > I'm creating root with 200Mbit
> > and parent with 10Mbit/1Mbit
> >
> >
> > class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 131072 rate 10Mbit ceil 12Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
> >  Sent 25443954 bytes 17155 pkts (dropped 0, overlimits 0)
> >  rate 143050bps 97pps backlog 25p
> >  lended: 17130 borrowed: 0 giants: 0
> >  tokens: 335 ctokens: -787
> >
> > class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
> >  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
> >  rate 2723bps 49pps
> >  lended: 0 borrowed: 0 giants: 0
> >  tokens: 8241 ctokens: 8241
> >
> > class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
> >  Sent 25406104 bytes 17130 pkts (dropped 0, overlimits 0)
> >  rate 144147bps 98pps
> >  lended: 0 borrowed: 0 giants: 0
> >  tokens: 8195 ctokens: 8195
> >
> > class htb 1:3500 parent 1:1 leaf 3500: prio 0 quantum 13107 rate 1Mbit ceil 1Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
> >  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
> >  rate 2739bps 49pps
> >  lended: 8712 borrowed: 0 giants: 0
> >  tokens: 12501 ctokens: -294
> >
> >
> > so, rate is 1,2Mbit
> > and what is meaning of negative ctokens?
> >
> > > >From data you send one can't decide what's bad. In any case
> > > and as I said before, behavioue changed a bit so that results
> > > can be a bit different a may need tc script changes.
> > >
> > > devik
> > >
> > > _______________________________________________
> > > LARTC mailing list / LARTC@mailman.ds9a.nl
> > > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
> >
> > --
> > Michael Vasilenko
> >
> >
> 
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

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

Re: [LARTC] htb2 -> htb3 problems

[devik]

remove cburst 1

-------------------------------
    Martin Devera aka devik
Linux kernel QoS/HTB maintainer
  http://luxik.cdi.cz/~devik/

On Tue, 6 Apr 2004, Michael Vasilenko wrote:

> devik (devik@cdi.cz) wrote:
> > I see you have cburst 0 ! It is not allowed
> > and is described in docs. What was commands to
> > create the classes ?
>
>
> /sbin/tc.3 qdisc add dev eth0 root handle 1:0 htb default 2500 r2q 100
> /sbin/tc.3 class add dev eth0 parent 1:0 classid 1:1 htb rate 200mbit ceil 200mbit quantum 200000
> /sbin/tc.3 class add dev eth0 parent 1:0 classid 1:2 htb rate 200mbit ceil 200mbit quantum 200000
> /sbin/tc.3 class add dev eth0 parent 1:1 classid 1:3500 htb rate 10Mbit ceil 10Mbit burst 2048b cburst 1
> /sbin/tc.3 qdisc add dev eth0 parent 1:3500 handle 3500: sfq perturb 5
> /sbin/tc.3 class add dev eth0 parent 1:2 classid 1:5500 htb rate 20Mbit ceil 20Mbit burst 2048b cburst 1 quantum 60000
> /sbin/tc.3 qdisc add dev eth0 parent 1:5500 handle 5500: sfq perturb 5
> /sbin/tc.3 filter add dev eth0 parent 1:0 protocol ip prio 50 handle 3500 fw classid 1:3500
> /sbin/tc.3 filter add dev eth0 parent 1:0 protocol ip prio 50 handle 5500 fw classid 1:5500
>
>
>
> tc.3 -V
> tc utility, iproute2-ss020116
>
> > On Tue, 6 Apr 2004, Michael Vasilenko wrote:
> >
> > > devik (devik@cdi.cz) wrote:
> > > > > > I suppose the drop you see is CPU bound ? Did you profiled it
> > > > > > again ?
> > > > >
> > > > > No, CPU is about times more in idle state in htb3, and
> > > > > 5x lowest data rate. AFAIK, htb3 scheduler is faster, so that is the
> > > > > reason of moving to htb3.
> > > >
> > > > ok then it is config issue probably. you should pin it down
> > > > to smallest possible number of classes for test (say up to
> > > > 5 classes) and then use tc -s show class ... to see internal
> > > > statistics. Look for classes with small (or negative)
> > > > tokens or ctokens - these are in throttling state and are
> > > > slowing throughtput - then think if it is ok ..
> > >
> > > ok
> > > I'm creating root with 200Mbit
> > > and parent with 10Mbit/1Mbit
> > >
> > >
> > > class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 131072 rate 10Mbit ceil 12Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
> > >  Sent 25443954 bytes 17155 pkts (dropped 0, overlimits 0)
> > >  rate 143050bps 97pps backlog 25p
> > >  lended: 17130 borrowed: 0 giants: 0
> > >  tokens: 335 ctokens: -787
> > >
> > > class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
> > >  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
> > >  rate 2723bps 49pps
> > >  lended: 0 borrowed: 0 giants: 0
> > >  tokens: 8241 ctokens: 8241
> > >
> > > class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
> > >  Sent 25406104 bytes 17130 pkts (dropped 0, overlimits 0)
> > >  rate 144147bps 98pps
> > >  lended: 0 borrowed: 0 giants: 0
> > >  tokens: 8195 ctokens: 8195
> > >
> > > class htb 1:3500 parent 1:1 leaf 3500: prio 0 quantum 13107 rate 1Mbit ceil 1Mbit burst 2Kb/8 mpu 0b cburst 0b/8 mpu 0b level 0
> > >  Sent 482570 bytes 8712 pkts (dropped 0, overlimits 0)
> > >  rate 2739bps 49pps
> > >  lended: 8712 borrowed: 0 giants: 0
> > >  tokens: 12501 ctokens: -294
> > >
> > >
> > > so, rate is 1,2Mbit
> > > and what is meaning of negative ctokens?
> > >
> > > > >From data you send one can't decide what's bad. In any case
> > > > and as I said before, behavioue changed a bit so that results
> > > > can be a bit different a may need tc script changes.
> > > >
> > > > devik
> > > >
> > > > _______________________________________________
> > > > LARTC mailing list / LARTC@mailman.ds9a.nl
> > > > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
> > >
> > > --
> > > Michael Vasilenko
> > >
> > >
> >
> > _______________________________________________
> > LARTC mailing list / LARTC@mailman.ds9a.nl
> > http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
>
> --
> Michael Vasilenko
>
>

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

Re: [LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

devik (devik@cdi.cz) wrote:
> remove cburst 1

thanks!


class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 65536 rate 5Mbit ceil 5Mbit burst 20Kb/8 mpu 0b cburst 8151b/8 mpu 0b level 0
 Sent 45107618 bytes 42138 pkts (dropped 0, overlimits 0)
 rate 608617bps 548pps backlog 40p
 lended: 42098 borrowed: 0 giants: 0
 tokens: 3908 ctokens: -11502

class htb 1:1 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
 Sent 369059 bytes 6711 pkts (dropped 0, overlimits 0)
 rate 5706bps 104pps
 lended: 0 borrowed: 0 giants: 0
 tokens: 8241 ctokens: 8241

class htb 1:2 root rate 200Mbit ceil 200Mbit burst 263690b/8 mpu 0b cburst 263690b/8 mpu 0b level 7
 Sent 45055458 bytes 42098 pkts (dropped 0, overlimits 0)
 rate 613755bps 549pps
 lended: 0 borrowed: 0 giants: 0
 tokens: 7822 ctokens: 7822

class htb 1:3500 parent 1:1 leaf 3500: prio 0 quantum 131072 rate 10Mbit ceil 10Mbit burst 20Kb/8 mpu 0b cburst 14704b/8 mpu 0b level 0
 Sent 369059 bytes 6711 pkts (dropped 0, overlimits 0)
 rate 5833bps 106pps
 lended: 6711 borrowed: 0 giants: 0
 tokens: 12713 ctokens: 9104

but I see only 330Kbytes/sec flow (half of 5Mbit rate)

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

Re: [LARTC] htb2 -> htb3 problems

[Martin Devera]

> > remove cburst 1
>
> thanks!
>
> class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 65536 rate 5Mbit
> ceil 5Mbit burst 20Kb/8 mpu 0b cburst 8151b/8 mpu 0b level 0
>  Sent 45107618 bytes 42138 pkts (dropped 0, overlimits 0)
>  rate 608617bps 548pps backlog 40p
>  lended: 42098 borrowed: 0 giants: 0
>  tokens: 3908 ctokens: -11502
>
> but I see only 330Kbytes/sec flow (half of 5Mbit rate)

From above I see rate 608617bps = 4.7MBit. Probably it will settle down
after some while. I suspect there is problem in your way of measuring the
rate. Also you can deliberately increase (c)burst to more than default,
say to 20kB - the larger the burst it the more insensitive to CPU load
it is ...

devik

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

Re: [LARTC] htb2 -> htb3 problems

[Michael Vasilenko]

Martin Devera (devik@cdi.cz) wrote:
> > > remove cburst 1
> >
> > thanks!
> >
> > class htb 1:5500 parent 1:2 leaf 5500: prio 0 quantum 65536 rate 5Mbit
> > ceil 5Mbit burst 20Kb/8 mpu 0b cburst 8151b/8 mpu 0b level 0
> >  Sent 45107618 bytes 42138 pkts (dropped 0, overlimits 0)
> >  rate 608617bps 548pps backlog 40p
> >  lended: 42098 borrowed: 0 giants: 0
> >  tokens: 3908 ctokens: -11502
> >
> > but I see only 330Kbytes/sec flow (half of 5Mbit rate)
> 
> >From above I see rate 608617bps = 4.7MBit. Probably it will settle down
> after some while. I suspect there is problem in your way of measuring the
> rate. Also you can deliberately increase (c)burst to more than default,
> say to 20kB - the larger the burst it the more insensitive to CPU load
> it is ...

OK, I removed all manual settings of burst/cburst, and it works just
fine. htb3 rocks! 30%/90% CPU Usage on 80Mbit stream with >500 classes
on htb3/htb2 (P4 2Ghz, 2.4.25 kernel with APIC, e1000 NIC)

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