comp.os.linux.networking

"buck" wrote in message
news:hbqts39fi7p60sgcq7sfu472hfn0pbmhbu@4ax.com...
> On Mon, 3 Mar 2008 12:42:21 -0800, "D. Stussy"
wrote:
> >Under the old way (using a "shaper#" pseudo-network-device), all one had
to
> >do was to attach it, set the desired speed/maximum bandwidth, then route
the
> >desired destination address(es) through it. The new way uses the "tc"
> >program from the iproute2 package. I have searched for examples on how
to
> >do this with "tc" and have found some, but am still unclear. The manual
> >page for "tc" doesn't seem to list all the permutations (or options) of
the
> >command, nor does it really explain what the different queueing
disciplines
> >are. I also don't fully follow what the "classes" are and why they're
> >needed. The filter subsystem makes sense but also isn't fully explained.
I
> >would have liked to have found an example of the old, depreciated
"shaper0"
> >interface and its comparable "tc" setup but such has eluded me.
> >
> >Where's a GOOD resource that explains ALL the workings of "tc"?
>
> My opinion: Where is good documentation for anything in Linux?
> Programmers are poor authors and good authors usually have no clue
> what a program really does.
>
> >The resources I printed so far are:
> >http://linux.die.net/man/8/tc-tbf
> >http://www.linuxdocs.org/HOWTOs/Adv-Routing-HOWTO-14.html
> >http://www.topwebhosts.org/tools/traffic-control.php
>
>http://www.linux.org/docs/ldp/howto/Adv-routing-HOWTO/lartc.ratelimit.singl
e.html
> >
> >None of these explain why I need to state the bandwidth for the main
device
> >in the qdisc line. If the value is really used, I have an added
> >complication of that I cannot necessarily determine the value that will
be
> >used for an ethernet interface (it's a 100Mbit, but the colocation
provider
> >sometimes attaches a slower ethernet segment (10Mbit or effective 20Mbit
as
> >10 full-duplex)). I could just follow the examples, but they're
meaningless
> >unless I understand what they're doing.
>
> I think you're looking in the wrong place. For simple rate limiting,
> check iptables.

Actually, using iptables doesn't do the job as I want the additional packets
queued or delayed (but still eventually sent), not dropped or rejected.

> TC has no decent docs. You have found what there is. Your situtation
> prevents the use of a queueing discipline because these only work when
> the available bitrate has an absolute maximum that never varies.

I have continued to explore and found that only the cbq qdisc had this
brain-dead arrangement. The htb and tbf qdiscs were much more reasonable,
and I think I figured out something that works for me using them. However,
getting all the parameters correctly stated took about an hour as for many
of them, the man pages dont' really cover them well. I probably have some
that are unnecessary - carried from the examples I found.

I really liked the reference to the "tc-filters" man page which doesn't even
exist!

What I needed to do was to rate-limit outbound packets to a particular
destination IP. That should have been obvious from my reference to the
depreciated "shaper0" virtual interface.