Skip to contentSkip to conversation

HTB statistics

LARTC maitri 6 participants 8 messages Apr 14, 2003 – Oct 1, 2007

lore
Share & tools
lore
#1 lore raw
Quoted 2 lines · maitri

i want to know what it means to have negative value of 'tokens' and 'ctokens' in the HTB statistics or token bucket filtering.

This is not so complicated. Plz read the docs at www.docum.org. Especially
http://www.docum.org/stef.coene/qos/tests/htb/burst/

Stef

Signature · 8 lines
[email protected]
 "Using Linux as bandwidth manager"
     http://www.docum.org/
     #lartc @ irc.oftc.net

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
#2 lore raw
In reply to maitri

Hi,
there is something I'm not able to understand about HTB statistics.
Why the qdisc show show how many packets are overlimits but in the classes there are always 0 packets overlimits or dropped?

I try to expain with a simple example. I create a qdisc htb with only a class and then i do a lot of traffic.

pc-innocenti:~# tc qdisc add dev eth0 parent root handle 1: htb default 10
pc-innocenti:~# tc class add dev eth0 parent 1:1 classid 1:10 htb rate
1kbit cei l 1kbit quantum 2000

The qdisc show that there are packets overlimits:

pc-innocenti:~# tc -s -d qdisc show
qdisc htb 1: dev eth0 r2q 10 default 10 direct_packets_stat 1145 ver 3.17
  Sent 339419 bytes 1533 pkts (dropped 0, overlimits 402)
  backlog 370p

The class instead always show 0 packet overlimits or dropped.

pc-innocenti:~# tc -s -d class show dev eth0
class htb 1:10 root prio 0 quantum 2000 rate 1000bit ceil 1000bit burst
1600b/8 mpu 0b overhead 0b cburst 1600b/8 mpu 0b overhead 0b level 0
  Sent 66307 bytes 480 pkts (dropped 0, overlimits 0)
  rate 912bit backlog 448p
  lended: 32 borrowed: 0 giants: 0
  tokens: -13238252 ctokens: -13238252
Signature · 19 lines
**********************************************************************
Marco Innocenti              Gruppo Infrastruttura e Sicurezza
CINECA                       phone:+39 0516171553 / fax:+39 0516132198
Via Magnanelli 6/3           e-mail: [email protected]
40033 Casalecchio di Reno    Bologna (Italia)
**********************************************************************


-- 
**********************************************************************
Marco Innocenti              Gruppo Infrastruttura e Sicurezza
CINECA                       phone:+39 0516171553 / fax:+39 0516132198
Via Magnanelli 6/3           e-mail: [email protected]
40033 Casalecchio di Reno    Bologna (Italia)
**********************************************************************
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
#3 lore raw
In reply to maitri
Quoted 4 lines · [email protected] wrote:

Hi,
there is something I'm not able to understand about HTB statistics.
Why the qdisc show show how many packets are overlimits but in the classes there are always 0 packets overlimits or dropped?

Overlimits is more a kernel thing then usefull as a traffic control figure and tc qdisc -s .. shows the root 1: htb but tc -s class doesn't.

You can see drops with tc -s class - but the default htb queue is too long in this test. Add a specific queue to the class 1:10 so you can specify length. eg.

tc qdisc add dev eth0 parent 1:10 handle 10: bfifo limit 64k

Quoted 13 lines

I try to expain with a simple example. I create a qdisc htb with only a class and then i do a lot of traffic.

pc-innocenti:~# tc qdisc add dev eth0 parent root handle 1: htb default 10
pc-innocenti:~# tc class add dev eth0 parent 1:1 classid 1:10 htb rate
1kbit cei l 1kbit quantum 2000

The qdisc show that there are packets overlimits:

pc-innocenti:~# tc -s -d qdisc show
qdisc htb 1: dev eth0 r2q 10 default 10 direct_packets_stat 1145 ver 3.17
 Sent 339419 bytes 1533 pkts (dropped 0, overlimits 402)
 backlog 370p
Quoted 10 lines

The class instead always show 0 packet overlimits or dropped.

pc-innocenti:~# tc -s -d class show dev eth0
class htb 1:10 root prio 0 quantum 2000 rate 1000bit ceil 1000bit burst
1600b/8 mpu 0b overhead 0b cburst 1600b/8 mpu 0b overhead 0b level 0
 Sent 66307 bytes 480 pkts (dropped 0, overlimits 0)
 rate 912bit backlog 448p
 lended: 32 borrowed: 0 giants: 0
 tokens: -13238252 ctokens: -13238252

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#4 lore raw
In reply to maitri
Quoted 9 lines · Andy Furniss
Quoted 3 lines

there is something I'm not able to understand about HTB statistics.
Why the qdisc show show how many packets are overlimits but in the classes there are always 0 packets overlimits or dropped?

Overlimits is more a kernel thing then usefull as a traffic control figure and tc qdisc -s .. shows the root 1: htb but tc -s class doesn't. You can see drops with tc -s class - but the default htb queue is too long in this test. Add a specific queue to the class 1:10 so you can specify length. eg.
tc qdisc add dev eth0 parent 1:10 handle 10: bfifo limit 64k

Thank you for your reply. So there is not a way to know that a specific class is actually shaping the traffic without attaching a qdisc to each class?

Signature · 10 lines
**********************************************************************
Marco Innocenti              Gruppo Infrastruttura e Sicurezza
CINECA                       phone:+39 0516171553 / fax:+39 0516132198
Via Magnanelli 6/3           e-mail: [email protected]
40033 Casalecchio di Reno    Bologna (Italia)
**********************************************************************
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
#5 lore raw
In reply to maitri

<snip>
Thank you for your reply. So there is not a way to know that a specific class is actually shaping the traffic without attaching a qdisc to each class?
</snip>

In the current implementation, tc does not report on default qdiscs. Therefore, if would like the ability to follow the amount of traffic flowing through a particular queue I recommend that you attach a PFIFO (the default).

-Ron

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#6 lore raw
In reply to maitri
Quoted 14 lines · [email protected] wrote:
Quoted 10 lines · Andy Furniss
Quoted 3 lines

there is something I'm not able to understand about HTB statistics.
Why the qdisc show show how many packets are overlimits but in the classes there are always 0 packets overlimits or dropped?

Overlimits is more a kernel thing then usefull as a traffic control figure and tc qdisc -s .. shows the root 1: htb but tc -s class doesn't. You can see drops with tc -s class - but the default htb queue is too long in this test. Add a specific queue to the class 1:10 so you can specify length. eg.
tc qdisc add dev eth0 parent 1:10 handle 10: bfifo limit 64k

Quoted 5 lines

Thank you for your reply. So there is not a way to know that a specific class is actually shaping the traffic without attaching a qdisc to each class?

Well the fact you are backlogged shows queue length - just default length htb uses in this test is so long it doesn't drop.

If you had more classes they would be shorter and drop.

Adding a queue just lets you choose length - you won't really see any better stats though you see more with tc -s qdisc.

You can also use tc -s class and look at the borrowed/lended counts and see what you can deduce from those.

Andy.

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#7 lore raw

[LARTC] htb statistics

In reply to maitri

Hi,

how long is the period over that the htb rate of a class will be estimated? on a site i read 10 sec but the update rate is 10 sec. i think the average rate determination is nearly 120 seconds?
What is the exact meaning of requeues? If one packet must be 5 times requeued this statistics value is 5? When will be a packet requeued because in my case this value is always 0...
Is overlimits a summary of dropped, requeued and delayed packets, because all dropped or requeued packets were previously marked as overlimit?
I have a htb root qdisc,three htb classes, and two leaf-sfq-qdiscs. The root class will shaping the traffic to 512kbit. So i meant that the overlimits value of this class will be increased if someone would send the data faster. but the overlimits will be counted only in the root htb qdisc. is there a reason for this behauvior?

Thanks in advance for your help!
Gerrit

____________________________________________________________________________________
Need a vacation? Get great deals
to amazing places on Yahoo! Travel.
http://travel.yahoo.com/
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc