Re: [LARTC] RE: Sluggish throughput with htb

[gypsy <gypsy@iswest.com>]

Andrew Kraslavsky wrote:
> 
> All,
> 
> Follow up to the problem I reported earlier:
> 
> > I have been using the following as a means of rate limiting
> > access to the Internet via eth0 (which connects to my cable
> > modem) and it was working great with my 2.4.20 kernel:
> >
> > tc qdisc del dev eth0 root
> > tc qdisc add dev eth0 root handle 1: htb default 1
> > tc class add dev eth0 parent 1: classid 1:1 htb rate 486kbit ceil 486kbit
> > tc qdisc add dev eth0 parent 1:1 handle 10: sfq perturb 10
> >
> > However, after recently updating to kernel version 2.6.15,
> > throughput on eth0 has become sluggish and it is only by
> > disabling the above that performance is restored.
> >
> > Also, I see a whole slew of the following messages in my log:
> >
> > qdisc_restart: Empty queue has non zero length 1
> 
> I see negative "tokens" and "ctokens" values listed when I do a "show" on the class of eth0:
> 
> $ tc -s -d class show dev eth0
> 
> class htb 1:1 root leaf 10: prio 0 quantum 6075 rate 486000bit ceil 486000bit burst 2206b/8 mpu 0b overhead 0b cburst 2206b/8 mpu 0b overhead 0b level 0

Why is the quantum so large?  Is this a gigabit NIC?

>  Sent 614100 bytes 578 pkts (dropped 0, overlimits 0)
>  rate 31312bit 3pps backlog 106p

This backlog and the 472 borrowed below are what are causing your
negatives.

>  lended: 472 borrowed: 0 giants: 0
>  tokens: -33271 ctokens: -33271
> 
> Are the "tokens" and "ctokens" values ever supposed to be negative?

Yes.

> If so, what does the negative value mean (e.g. borrowed or lent tokens?).

I don't know.  Read the HTB docs.

  If not, is this indicative of a problem that is at all familiar to
anyone?
> 
> Thanks,
> 
> - Andrew Kraslavsky

My guesses are:
1) The clock used by HTB is too slow.
2) This is a gigabit NIC not correctly set up.
--
gypsy
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc