Re: netem and hierarchical ingress traffic shaping

[Eric Dumazet <eric.dumazet@xxxxxxxxx>]

Le vendredi 23 décembre 2011 à 18:33 +0100, Eric Dumazet a écrit :
> Le dimanche 18 décembre 2011 à 11:55 -0800, Stephen Hemminger a écrit :
> > On Sun, 18 Dec 2011 00:12:12 -0500
> > "John A. Sullivan III" <jsullivan@xxxxxxxxxxxxxxxxxxx> wrote:
> > 
> > > Since netem appears to be classless, we realized we would need to
> > > replace the SFQ on each leaf with netem which we really didn't want to
> > > do - not only to not lose SFQ but because we didn't want to maintain the
> > > netem parameters on each leaf.  So, we activated our ifb1 interface,
> > > placed netem on it and redirected all the egress traffic to ifb1.  Taht
> > > worked fine.
> > 
> > Current versions of netem can take one class.
> > --
> 
> Hmm, I can see that (commit 10f6dfcfde884441) 
> 
> But it wont work very well, it assumes qdisc uses a single queue if
> netem reordering is requested :
> 
> (__skb_queue_head(&q->qdisc->q, skb))
> 
> We should allow reordering if netem queue is changed from tfifo, only if
> new qdisc is compatible with __skb_queue_head(&q->qdisc->q, skb)
> 
> (maybe providing a new ->queue_at_head() new ops)

Also, child qdisc must not scratch skb->cb[], since netem stores
time_to_send in it.

I guess nobody actually tried this netem mis-feature :(

I'll send a fix.



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html