Gigabit Linux Routers

Joe Greco jgreco at ns.sol.net
Fri Dec 19 23:40:16 UTC 2008


> Dear Joe,
> > Yes, but the point was that the feature was listed as "simple traffic
> > shaping."  You can do *complicated* traffic shaping too, which was the
> > reason I commented on that.  Usually the ability to do complicated
> > traffic shaping means you can do simple traffic shaping too.  ;-)
> 
> with linux?
> really?

Reread the message...  the text was in reply to a discussion of FreeBSD
features.  And, yes, really.  ipfw, pf, and ipf solutions are all
trivially available, giving you a selection of rule types and altq or
dummynet shaping options that can be tied into extremely flexible rules.

> > Mmm, generally, it looks to me like it works, but the above is the
> > entirety of my testing, so I could easily be wrong.
> 
> you have ospf between this 2 boxes?

Yes.  vlan20 is OSPF-enabled, as a matter of fact, and both routers are
on it.  The goal was to see if I could get a network that was smart enough
for both OSPF-enabled hosts ("no static gateway needs to be config'ed")
and non-OSPF hosts (CARP as default gateway).

> show me them routing table.
> do a failover and show the routing table again,

I did that experiment below.  I didn't grab snapshots of the routing table
at the time, but I described the effect.  Essentially, upon downing of the
interface, the local link via the vlan20 interface went away, and was 
promptly replaced by the OSPF route (generally good/desirable).  Further
discussion was in my previous message.

> >> *) carp is i bound, carp-dev line openbsd is in development
> >>  	(not shure if already stable)
> >
> > You mean inbound?  Well, yes.  That's reasonably practical.  It isn't
> > entirely clear what other paradigms would look like (i.e. if the host
> > system didn't have a native address on the wire), though several ideas
> > spring to mind.
> >
> > Am I correct in assuming that you mean to have no native interface on
> > the network in question, and only a CARP interface?  Or am I reading in
> > between the lines incorrectly?
> 
> only carp-int has the ip's.

Really?  Interesting.  I'm trying to think of how that would be configured.
How does the system identify which ethernet interface to use, or is this
something that's specific to Linux?

> >> *) if carp switch over:
> >>  	t=0: A is master, has route 192.168.0.1/24
> >>  	     B has route 192.168.0.1/24 via ospf
> >>  	t=1: A goes down, route disappear (need linkstate in ospf)
> >>  	t=2: B carp takes over 192.168.0.1/24
> >>  		B can not add 192.168.0.1/24 route as it is still
> >>  		known via ospf
> >>  	t=3: B gets update to remove route 192.168.0.1/24 via ospf
> >>  	t=4: 192.168.0.1/24 route has disappeared, failover broken.
> >>
> >> with ucarp, some special scripts and source code changed I was able
> >> to handle this situation, but not with carp and ospf (at least at
> >> freebsd 6.3)
> >
> > I agree that this is a problematic scenario.  FreeBSD 5.* and 6.* are
> > pretty worthless to us, so we've pretty much jumped from 4 to 7, and
> > so my knowledge of the networking improvements in between are limited.
> 
> I have not yet tested freebsd 7, as the multicast kernel interface 
> changed and quagge ospf breaked. also I need(ed) a stable platform.

I'm aware of the Quagga OSPF issues, having grinched about them a number
of times in various places.  For what it is worth, there's a patch that
appears to work, but which was thought to not really be a "correct" fix.
Several people, including us, however, are using it with apparent success.
 
> > Under FreeBSD 4, there is indeed a great deal of pain associated with
> > routes coming in via a routing protocol that are also theoretically
> > available on a directly-attached interface.
> >
> > I just tried downing rtr1: vlan20 on the above (which is FreeBSD 7,
> > obviously) and from rtr1's PoV the network did move correctly to an
> > alternate route via OSPF, but upon re-enabling the vlan20 interface,
> > the OSPF route remained.  Now, it seemed to "all work again" when I
> > did the following:
> 
> yes, thats the problem.
> 
> > # ifconfig vlan20 up
> > # route delete -net 206.55.68.192
> > # ifconfig vlan20 inet 206.55.68.195 netmask 0xffffffe0
> 
> I have changed ucarp todo so, but you also need
> gratious arp and such stuff to get a real, flawless failover.

Don't know.
 
> > which re-established the local link.  That's not ideal, but it is a lot
> > better than FreeBSD 4, where things were just breaking all over if you
> > did "strange" things like this.
> >
> > For most important things around here, we use OSPF with stub routes so
> > the failure of a particular ethernet is not necessarily of great concern,
> > but it would be nice to see things like this know how to DTRT.
> 
> DTRT?

Do the right thing.

... JG
-- 
Joe Greco - sol.net Network Services - Milwaukee, WI - http://www.sol.net
"We call it the 'one bite at the apple' rule. Give me one chance [and] then I
won't contact you again." - Direct Marketing Ass'n position on e-mail spam(CNN)
With 24 million small businesses in the US alone, that's way too many apples.




More information about the NANOG mailing list