All of lore.kernel.org
 help / color / mirror / Atom feed
* [LARTC] SFQ latency
@ 2004-05-19  9:21 Tomas Simonaitis
  2004-05-19 10:58 ` Andy Furniss
  2004-05-19 13:23 ` Tomas Simonaitis
  0 siblings, 2 replies; 3+ messages in thread
From: Tomas Simonaitis @ 2004-05-19  9:21 UTC (permalink / raw)
  To: lartc

Anyone notices that using sfq qdisc adds latency compared to pfifo?

With "sfq perturb 10" ping latency is around 100-120ms, while with "pfifo 
limit 3" drops to ~10-30ms (qdiscs belong to htb class).

I'm missing something or is this expected?

{
$TC_CLASS parent 1:1 classid 1:20 htb prio 0 rate 900kbit ceil 1000kbit burst 
7k
$TC_QDISC parent 1:20 handle 200: {pfifo limit 3} | {sfq perturb 10}
}
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [LARTC] SFQ latency
  2004-05-19  9:21 [LARTC] SFQ latency Tomas Simonaitis
@ 2004-05-19 10:58 ` Andy Furniss
  2004-05-19 13:23 ` Tomas Simonaitis
  1 sibling, 0 replies; 3+ messages in thread
From: Andy Furniss @ 2004-05-19 10:58 UTC (permalink / raw)
  To: lartc

Tomas Simonaitis wrote:
> Anyone notices that using sfq qdisc adds latency compared to pfifo?
> 
> With "sfq perturb 10" ping latency is around 100-120ms, while with "pfifo 
> limit 3" drops to ~10-30ms (qdiscs belong to htb class).
> 
> I'm missing something or is this expected?
> 
> {
> $TC_CLASS parent 1:1 classid 1:20 htb prio 0 rate 900kbit ceil 1000kbit burst 
> 7k
> $TC_QDISC parent 1:20 handle 200: {pfifo limit 3} | {sfq perturb 10}
> }

There shouldn't really be a queue as such (you still choose one) for 
your interactive traffic - if you really care about latency HTB should 
have a class with a high (low number) prio and a high rate (> expected 
traffic), so other traffic has to borrow.

If you do end up queueing your interractive, then pfifo limit 3 means 
there will only be 2 packets in front of the ping (I suppose bfifo would 
be nicer as a fall back - it will penalise big packets more, WRT drops). 
SFQ default qlen is 128, though the way it works means the delay will 
depend on how many other connections are queued.


Andy.


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

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [LARTC] SFQ latency
  2004-05-19  9:21 [LARTC] SFQ latency Tomas Simonaitis
  2004-05-19 10:58 ` Andy Furniss
@ 2004-05-19 13:23 ` Tomas Simonaitis
  1 sibling, 0 replies; 3+ messages in thread
From: Tomas Simonaitis @ 2004-05-19 13:23 UTC (permalink / raw)
  To: lartc

Thanks Andy.

{P.S. 1:20 is class for interactive traffic, i skipped other}
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2004-05-19 13:23 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-05-19  9:21 [LARTC] SFQ latency Tomas Simonaitis
2004-05-19 10:58 ` Andy Furniss
2004-05-19 13:23 ` Tomas Simonaitis

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.