Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 27 Oct 2000 00:58:10 -0600 (MDT)
From:      Nick Rogness <nick@rapidnet.com>
To:        "Richard A. Steenbergen" <ras@e-gerbil.net>
Cc:        Bakul Shah <bakul@torrentnet.com>, "Ron 'The InSaNe One' Rosson" <insane@lunatic.oneinsane.net>, freebsd-net@FreeBSD.ORG
Subject:   Re: Multihomed Routing 
Message-ID:  <Pine.BSF.4.21.0010270034240.18689-100000@rapidnet.com>
In-Reply-To: <Pine.BSF.4.21.0010270205210.10623-100000@overlord.e-gerbil.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 27 Oct 2000, Richard A. Steenbergen wrote:

> On Thu, 26 Oct 2000, Nick Rogness wrote:
> 
> > 	You are assuming that the network that machine1 lies on has only 1
> > 	machine on it.  What happens when you add 2 more machines to that
> > 	network?  Now, router1 has to handle redirects for all of those
> > 	machines as well.
> > 
> > 	1 machine = 200 redirects
> > 	2 machines = 400 redirects (200 for machine1 & 200 for machine2)
> > 	3 machines = 600 redirects
> 
> In practice this is beyond silly (and most hosts should probably not be
> honoring redirects for security reasons). If reliability is that important
> to you, you should have routers which support a redundancy protocol. This
> will scale many orders of magnitude further then informing every host of
> available routes, especially as the number of hosts and the number of
> routes increase.

	That is the main reason you use dynamic routing on the
	hosts.  To keep routing tables simple on the hosts.
	Carrying subnet routes instead of host routes.	

	Multiple paths to multiple networks can become a nightmare without
	it ;-)

	The only reason I mention this is because I have had to deal with
	this issue in the past.  I've seen routers load to 30% just
	handling all of the ICMP redirects.  This solution eliminates
	that and all you have to do is run a simple routing daemon on the
	machines.  Set it up once...let it do the rest.

> 
> The only advantages of pushing the routing decision down to the host is
> A) load balancing, and B) the asthetic value of one less hop if the best
> exit is not available on the router you ended up hitting.
> 

	Yes.  What happens when you have multiple networks with multiple
	paths?  Your default router handles redirects for all of those
	networks.


> For point A, if you have two NICs and a legitimate need to balance across
> them at an IP layer, go for it.
> 
> For point B, I would venture to bet that the local communication between
> two routers sitting beside each other is far more reliable then trying to
> push a full routing table down to every host. :P
> 

	Don't get me wrong.  IMHO, I believe the routers should do most of
	the work. But on a large network this is sometimes not
	doable...because of design flaws or whatever.

> And if you design your network correctly many of these become non-issues.
> 

	There are several design reasons why you can't just make things as
	simple as you want them to be.  Geographical, financial,
	political, etc.  Limiting your design by not looking at all
	"angles" is ridiculous[spelling?].  I never once said this is the
	only way to do it...but it is an option.

	The point of the message was to make clear that there is other
	options with FreeBSD.  Not just always throwing in more routers
	and switches and adding net cards to machines when it can be done
	with the facilities provided to you.  Which [IMO] is what the
	FreeBSD project is all about.

	I hear you loud and clear though ;-p


Nick Rogness
- Drive defensively.  Buy a tank.




To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-net" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0010270034240.18689-100000>