Re: [LARTC] Some questions concerning IPtables (& IMQ/SFQ)

[Patrick McHardy <kaber@trash.net>]

Hi !
I guess i have to clarify this.

> - From Patricks' IMQ-page:
> ----
> SFQ is very useful as a leaf qdisc. But by default, its internal queue
> length is 128 which is too much for small classes or even for
> not-so-fast links. Changing SFQ_DEPTH in net/sched/ sch_sfq.c to about
> 10-20 results in flows responding much faster to bandwidth changes.
> ----
> 
> Is that ment for SFQ in general or only in conjunction with IMQ?

It was meant in conjunction with IMQ and ingress shaping.
The text is a bit misleading, i'm not sure if it is right.
My thought was:
One connection, 1 mbit rate, avg. pkt size 1480 = ~88pkt/sec.
To fill up entire SFQ_DEPTH of 128 it would take 1.45 seconds.
So if less bandwidth was available for this connection it would take at 
least this long + 1/2 rtt until package drops become noticeable for the 
remote side. This is probably not true as SFQ drops a (not necessary 
the last one!) packet from its longest slot if if queue becomes full.
I had a noticable effect in combination with imq but since then changed
the point at which imq tries to dequeue packets and in the mean time, i'm 
using it without changed SFQ_DEPTH.
I saw the same advice on another page, but like said before, i'm not sure 
if it is true. I'm going to think about it again as soon as my head clears 
up ;)

Bye,
Patrick

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