general network theory

Hi,

I have some general routing questions. On the network shown below I am
attempting to provide some level of reliability should a switch fail.

+----------+ +----------+
| isp feed | | isp feed |
+----+-----+ +----+-----+
| |
| |
+----+------+ +-----+-----+
| gw router | | gw router |
+--+-----+--+ +--+-----+--+
| a | b | c | d
| +---------+ | |
| | +---------+ |
| | | |
| +-------------+ |
| +-------------+ | |
| | | |
| | | |
+--+-----+---+ +--+-----+--+
| switch 3 | | switch 4 |
+--+-----+---+ +--+-----+--+
| | | |
| +--- < web servers > ---+ |
| |
| |
| |
+--+--------+ +--------+--+
| gw router | | gw router |
+--+-----+--+ +--+-----+--+
| e | b | c | d
| +---------+ | |
| | +---------+ |
| | | |
| +-------------+ |
| +-------------+ | |
| | | |
| | | |
+--+-----+---+ +--+-----+--+
| switch 1 | | switch 2 |
+-----+------+ +-----+-----+
| |
+------- < databases > -------+

Both the databases and the web servers have two network cards, in case
one network card fails. This leads to the problem of which card is used
with a default route.

Has anyone setup networks with hosts that exist within two networks.
How does the OS routing table respond to packets in this way? Surely
the source IP will match the default route, then be end up reaching the
source via a different link, and then fail to match the source's state
table.

--
Regards, Ed :: http://www.gnunix.net
just another python hacker
Visual Basic, much like generic beer and America's Funniest Home
Videos is an enabling technology for stupid people.

On Thu, 13 Jul 2006, in the Usenet newsgroup comp.os.linux.networking, in
article <(E-Mail Removed)>, ed wrote:

>I have some general routing questions. On the network shown below I am
>attempting to provide some level of reliability should a switch fail.

Hit google, and look for the "High Availability HOWTO"

130414 Dec 25 1998 High-Availability-HOWTO.tar.gz

That _was_ on Sunsite, but I have no idea where it's hiding now. (The
'linux-ha' directory is no longer on ibiblio.org.) The second document
you want is the Advanced Routing HOWTO

297491 Sep 4 2003 Adv-Routing-HOWTO

Do remember that the more complicated you make things to "improve
reliability", you are also creating more places for failure to occur. You
really want to put some numbers on the chance of something failing, and
what having some form of redundancy gains/costs, verses the increase in
the probability that a failure will occur caused by the multiple chances
you've provided. (Why do you think the majority of airliners being sold
today have two engines, not three or four?)

>Both the databases and the web servers have two network cards, in case
>one network card fails. This leads to the problem of which card is used
>with a default route.

See the documents above. Be careful - the word 'default' has different
meanings in several places. As far as Linux distributions are concerned,
the 'default route' is the 'route to the world' (more correctly, the
route to use when NO other route leads to the destination). You seem
to be using it in the sense 'route to use when more than one exists'.
That kind of confusion happens frequently, and the results tend to be
exactly opposite from each other.

>Has anyone setup networks with hosts that exist within two networks.

It's attempted all the time. Sometimes, people even get it to work.

Old guy