[LARTC] tc counters "problem"

[LARTC] tc counters "problem"

[Francisco]

Hi, I'm using tc and HTB to shape my outgoing ADSL traffic. I was trying to 
make some graphs on the classes by meassuring the "sent bytes" of each class 
using rrdtool to store the data (as kbps after conversion). I expected that 
meassuring the root class I would get values similar that the ones I get 
measuring the interface counters but they differ by a large amount.
Is there something obvious I'm missing here?, is this approach correct?
Thank you.

Francisco

kaori ~ # tc -s class show dev ppp0
class htb 1:1 root rate 200000bit ceil 200000bit burst 6Kb/8 mpu 0b overhead 
0b cburst 1699b/8 mpu 0b overhead 0b level 7
 Sent 378286130 bytes 2185057 pkt (dropped 0, overlimits 0 requeues 0)
 rate 182312bit 29pps backlog 0b 0p requeues 0
 lended: 0 borrowed: 0 giants: 0
 tokens: 168756 ctokens: -13270

class htb 1:10 parent 1:1 leaf 10: prio 1 rate 200000bit ceil 200000bit burst 
6Kb/8 mpu 0b overhead 0b cburst 1699b/8 mpu 0b overhead 0b level 0
 Sent 115548283 bytes 1520567 pkt (dropped 0, overlimits 0 requeues 0)
 rate 1848bit 4pps backlog 0b 0p requeues 0
 lended: 1520567 borrowed: 0 giants: 0
 tokens: 250020 ctokens: 67994

class htb 1:20 parent 1:1 leaf 20: prio 2 rate 180000bit ceil 180000bit burst 
6Kb/8 mpu 0b overhead 0b cburst 1689b/8 mpu 0b overhead 0b level 0
 Sent 262755089 bytes 664505 pkt (dropped 0, overlimits 0 requeues 0)
 rate 180456bit 25pps backlog 0b 15p requeues 0
 lended: 664490 borrowed: 0 giants: 0
 tokens: 113243 ctokens: -89463

class htb 1:30 parent 1:1 leaf 30: prio 3 rate 1000bit ceil 120000bit burst 
6Kb/8 mpu 0b overhead 0b cburst 1659b/8 mpu 0b overhead 0b level 0
 Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
 rate 0bit 0pps backlog 0b 0p requeues 0
 lended: 0 borrowed: 0 giants: 0
 tokens: 50331648 ctokens: 113321

class htb 1:40 parent 1:1 leaf 40: prio 4 rate 1000bit ceil 60000bit burst 
6Kb/8 mpu 0b overhead 0b cburst 1629b/8 mpu 0b overhead 0b level 0
 Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
 rate 0bit 0pps backlog 0b 0p requeues 0
 lended: 0 borrowed: 0 giants: 0
 tokens: 50331648 ctokens: 222548
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] tc counters "problem"

[Andreas Klauer]

On Sun, Apr 09, 2006 at 04:24:20PM -0300, Francisco wrote:
> I expected that meassuring the root class I would get values similar that 
> the ones I get measuring the interface counters but they differ by a 
> large amount.

The statistics you posted seem to be fine. Which interface counters are 
you talking about? tc starts counting only after the qdiscs/classes were 
created, so if you have a separate count somewhere which started counting 
some other time, the difference will of course be huge (as far as total 
packet count is concerned).

Regards
Andreas Klauer
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] tc counters "problem"

[Francisco]

The interface that I'm using is ppp0, the classes/qdiscs are created with the 
interface (from /etc/ppp/ip-up file) but that doesn't matter anyway, the way 
the two graphics are created is:
at time 0 look the total amount of bytes/bits transfered
at time 5 (mins) I look the amount of bytes/bits transfered and do: (bits(t=5) 
- bits(t=0))/(300secs) = bps (similar with bytes)
and repeat every five minutes, this is how MRTG graphs.
So it really doesn't matter if the counters are different, what matters is the 
difference between time x and time y.
The interface counters and the tc counters are fetched by SNMP and I'm 100% 
sure that my interface graph is accurate, I just don't understand why there 
is so much difference with the tc ones.

On the other hand, I just made a test with a file transfer resetting all 
counters:
file: 14675760 lego.pdf

iptables:
Chain OUTPUT (policy ACCEPT 478208 packets, 1617032905 bytes)
    pkts      bytes target     prot opt in     out     source             
destination
   10725 15568354 MARK       tcp  --  any    ppp0    anywhere             
anywhere            tcp spt:http MARK set 0x1e

tc:
class htb 1:30 parent 1:1 leaf 30: prio 3 rate 1000bit ceil 120000bit burst 
6Kb/8 mpu 0b overhead 0b cburst 1659b/8 mpu 0b overhead 0b level 0
 Sent 15568302 bytes 10724 pkt (dropped 0, overlimits 0 requeues 0)
 rate 112bit 0pps backlog 0b 0p requeues 0
 lended: 99 borrowed: 10625 giants: 0
 tokens: -13832192 ctokens: -671


All this seems file, the bytes match, etc. So I'm not sure what I'm doing 
wrong. I'll keep investigating, I just wanted to know if any of you saw 
something wrong with my reasoning or if there were known issues with the 
counters.

Francisco




El Domingo, 9 de Abril de 2006 16:50, Andreas Klauer escribió:
> On Sun, Apr 09, 2006 at 04:24:20PM -0300, Francisco wrote:
> > I expected that meassuring the root class I would get values similar that
> > the ones I get measuring the interface counters but they differ by a
> > large amount.
>
> The statistics you posted seem to be fine. Which interface counters are
> you talking about? tc starts counting only after the qdiscs/classes were
> created, so if you have a separate count somewhere which started counting
> some other time, the difference will of course be huge (as far as total
> packet count is concerned).
>
> Regards
> Andreas Klauer
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc