Date: Wed, 22 Feb 2006 15:18:41 -0800 From: "Jon Simola" <jon@abccomm.com> To: "Christopher McGee" <chris@xecu.net> Cc: freebsd-pf@freebsd.org Subject: Re: Hfsc configuration problems Message-ID: <8eea04080602221518x34da8bds88b713282ba52c74@mail.gmail.com> In-Reply-To: <43FCDEE7.2010802@xecu.net> References: <43FC9F63.5070009@xecu.net> <8eea04080602220957v46f9d11ev2544e8cbe893365d@mail.gmail.com> <43FCA7B8.3090300@xecu.net> <55e8a96c0602221042re25f819g1e3815384c022590@mail.gmail.com> <43FCB645.5000508@xecu.net> <8eea04080602221157h18555b9bxc2719b5a12f7362a@mail.gmail.com> <43FCDEE7.2010802@xecu.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2/22/06, Christopher McGee <chris@xecu.net> wrote:
> Jon Simola wrote:
>
> >On 2/22/06, Christopher McGee <chris@xecu.net> wrote:
> >
> >
> >
> >>I might be going about this the wrong way, but, this is ultimately what
> >>I'm trying to do. One queue has guaranteed 3Mb, another has a
> >>guaranteed 4Mb, another has 3Mb guarantee, which leaves about 90Mb as a
> >>pool for all of them. If they are backlogged, I want the first 2 queue=
s
> >>to be able to utilize the entire 90Mb, and the 3rd queue should be able
> >>to utilize about 25Mb of it. This is a simplified example of what I
> >>sent earlier. But if I can do this, I can accomplish what I'm trying t=
o
> >>and build off it later.
> >>
> >>
> >
> >If you're guaranteeing bandwidth allocation, then you want realtime,
> >and probably avoid the bandwidth declarations which are not quite the
> >same. HFSC is not trivial to get your head wrapped around and is
> >poorly documented because of that. It took me reading the man pages
> >and the PF guide several times over a couple months to get it
> >together.
> >
> >Another example for you to peruse:
> >
> >queue q_dmz hfsc(ecn realtime 50Mb upperlimit 90Mb) \
> > { q_dmz_h, q_dmz_hb, q_dmz_l, q_dmz_lb }
> >queue q_dmz_h hfsc(realtime 10% upperlimit 90%) priority 7
> >queue q_dmz_hb hfsc(realtime 20% upperlimit 90%) priority 4
> >queue q_dmz_l hfsc(realtime 5% upperlimit 90%) priority 3
> >queue q_dmz_lb hfsc(realtime 10% upperlimit 90%) priority 0
> >
> >Leave out the linkshare and bandwidth, just use realtime and
> >upperlimit. And the priority of the queues matters, in the above each
> >of the queues can go as high as 81Mb (90% of 90Mb) but if more than
> >one tries to go above 45Mb, the one with the higher priority gets
> >first chance at available bandwidth. Linkshare is another override; in
> >the above it is easily possible that the q_dmz_lb queue will get quite
> >backlogged as it gets last chance, adding linkshare would allow it to
> >bypass the priorities of the other queues. You may not want to even
> >use priorities, using just realtime and upperlimit is probably a lot
> >easier for your simplified example.
> >
> >Using the service curves is even more complex. This is all based on my
> >experience and research, so it may not be correct, but it's the
> >explanation that I use.
> >
> >--
> >Jon Simola
> >Systems Administrator
> >ABC Communications
> >
> >
> This information is very helpful. Here is my modified configuration,
> unfortunately it still gets the same errors:
>
> altq on $ext_if bandwidth 100Mb hfsc queue { high_pri, med_pri, junk }
> queue high_pri hfsc(realtime 128Kb upperlimit 500Kb red) priority 7
> queue med_pri hfsc(realtime 5Mb upperlimit 99% default ecn red) priority =
5
> queue junk hfsc(upperlimit 95% red)
Yet this compiles fine on my 5.3-STABLE box:
# pfctl -vvnAf test.pf
altq on em0 hfsc bandwidth 100Mb tbrsize 12000 queue { high_pri med_pri jun=
k }
queue high_pri bandwidth 1% priority 7 hfsc( red realtime 128Kb
upperlimit 500Kb )
queue med_pri bandwidth 1% priority 5 hfsc( red ecn default realtime
5Mb upperlimit 99% )
queue junk bandwidth 1% hfsc( red upperlimit 95% )
And a related sillier config on an OpenBSD 3.8 box (bandwidth 0%):
# pfctl -vvnAf test.pf
altq on em0 hfsc bandwidth 100Mb tbrsize 12000 queue { high_pri med_pri jun=
k }
queue high_pri bandwidth 0 b priority 7 hfsc( red realtime 128Kb
upperlimit 500Kb )
queue med_pri bandwidth 0 b priority 5 hfsc( red ecn default realtime
5Mb upperlimit 99% )
queue junk bandwidth 0 b hfsc( red upperlimit 95% )
> Just to clarify what I'm trying to do... I'm trying to guarantee 128K
> for high_pri stuff(ssh, carp, etc..) and allow up to 500Kfor it. Other
> services, like web traffic etc.. falls in med_pri, which gets 5Mb all
> the time but can use up to 99Mb. All other traffic, like cvsup, ftp,
> etc... will get no guarantees, but can use up to 95Mb if it's
> available. From all my reading, this seems correct but I still get the
> parent sc errors.
For some reason adding bandwidth lets it compile. I've got no idea
what that actually will do (non-sensical numbers). If you use some
sensible numbers for bandwidth and specify the realtime/upperlimit for
hfsc, I'd certainly be interested to hear how it goes for you.
--
Jon Simola
Systems Administrator
ABC Communications
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?8eea04080602221518x34da8bds88b713282ba52c74>