Date: Sun, 29 Jun 2008 17:43:14 +0200 From: Max Laier <max@love2party.net> To: .@babolo.ru Cc: freebsd-net@freebsd.org, Giulio Ferro <auryn@zirakzigil.org>, Alexandre Biancalana <biancalana@gmail.com> Subject: Re: altq on vlan Message-ID: <200806291743.15021.max@love2party.net> In-Reply-To: <1214651667.267043.71931.nullmailer@cicuta.babolo.ru> References: <1214651667.267043.71931.nullmailer@cicuta.babolo.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
On Saturday 28 June 2008 13:14:27 .@babolo.ru wrote:
> [ Charset ISO-8859-1 unsupported, converting... ]
>
> > On Friday 27 June 2008 18:57:59 Alexandre Biancalana wrote:
> > > On 6/27/08, Max Laier <max@love2party.net> wrote:
> > > > You don't need a patch at all. What you do is: Queue on the
> > > > physical interface, classify on the vlan interface. It is broken
> > > > to allow ALTQ on a virtual interface if you can do it otherwise.
> > > >
> > > > in pf.conf speak:
> > > >
> > > > If you have "ifconfig vlanX vlandev bge0 ..."
> > > >
> > > > altq on bge0 .... queue { vlan0, vlan1, ... }
> > > > queue vlan0 ... { vlan0_foo, vlan0_bar, ... }
> > > > queue vlan0_foo
> > > > queue vlan0_bar
> > > > ...
> > > >
> > > > pass on vlanX ... queue vlanX_foobar
> > > >
> > > > And there you go. No patch - whatsoever - required here.
> > >
> > > But the patch simplify the cases where you need one queue per vlan.
> >
> > NO! It is just wrong! There is no relation between vlan queues on
> > the same physical interface and thus you can't guarantee anything!
> > Can we please stop with this nonsense and not bring up the patch
> > every other month.
>
> Remember vlan anoter end.
>
> Vlan queues on the same physical interface has sense.
>
> Let see typical vlan use:
> +--------+ 100M untagged vlan1
> | |--------------..
> +---------+ | | 100M untagged vlan2
> 1G | | 1G tagged | |----------------
> --------+ FreeBSD +------------+ switch | 100M untagged vlan3
> | | | |--------------..
> +---------+ | | 100M untagged vlanN
> | |---------------
> +--------+
>
> There is noting interesting in common queue on 1G physical interface,
> the only right queues are that on vlans when number of
> vlans < 10.
>
> More of that, sum traffic on 1G tagged intervace is limited
> by incoming traffic from 1G external interface and
> so common queue on 1G tagged interface is not
> interesting even when number of vlans > 10.
Sorry, but you are completely off track here. If you use one queue per
vlan one vlan can easily DoS the rest, because once a packet has passed
the queue in the vlan it falls into a common queue with all the others
and - as you correctly point out - there is no guarantee that a 1G
interface can really sent at 1G all the time. The vlan queues, however,
will not get any feedback from the parent about it's real send speed.
E.g. a vlan sending *a lot* of tiny packets will dominate the 1G link and
thus DoS any other vlan that sends big packets. This you can prevent
with a common queue.
Now please ... let this die, it's stupid!
--
/"\ Best regards, | mlaier@freebsd.org
\ / Max Laier | ICQ #67774661
X http://pf4freebsd.love2party.net/ | mlaier@EFnet
/ \ ASCII Ribbon Campaign | Against HTML Mail and News
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200806291743.15021.max>