In your altq example you're not actually saturating the interface,
you've limited the bandwidth to 200Kb.
There is no non-altq bandwidth limitation on interfaces (yet), so when
using prio queueing you probably ARE saturating the interface, hence the
poor performance.
On Thu, Jul 21, 2011 at 10:56:42PM +0200, Christopher Zimmermann wrote:
> Hi,
>
> for me this new prio queue queueing just doesn't work on pppoe(4).
>
> I tested this now three times switching between the new prio and old
> altq/priq back and forth. With altq/priq I can use priorized ssh very
> smoothly while saturating my uplink with a 'nc foo.bar </dev/zero'.
> With the new prio I get echo delays of several seconds.
>
> Part of my pf.conf:
>
> [...]
>
> altq on pppoe0 priq bandwidth 200Kb queue { voip, lowdelay, default, bulk }
> queue voip priority 7 qlimit 3 priq
> queue lowdelay priority 6 qlimit 5 priq
> queue default priority 3 qlimit 30 priq(default)
> queue bulk priority 1 qlimit 50 priq
>
> [...]
>
> pass out proto tcp to port ssh queue(bulk,lowdelay)
>
> [...]
>
>
> With prio I simply replaced the queue(...) by prio(...) statements and
> the same numbers I used for priority in altq/priq:
> queue(bulk,lowdelay) -> prio(1,6).
>
>
> I don't know much about OpenBSDs network stack, but could it be that
> some lower level kernel driver (maybe pppoe(4)) has a queue which is
> dropping packets? Why else would altq/priq need the bandwidth
> parameter?!?
>
>
> Christopher
>
--
