Haim Ashkenazi wrote:
> Hi
> 
> I just came across this article:
> http://enterprise.linux.com/article.pl?sid=05/04/10/2132252&from=rss
> 
> It seems that I was doing things the hard way (using iproute with
> metrics). the one thing I don't understand is how to prefer one route
> over the other (e.g. when one connection is faster).

	I believe the author is counting on a total failure of eth0 to drop the 
primary route. Once eth0 goes down that route is withdrawn from the 
routing table leaving only the eth1 route. Linux without turning on IP: 
advanced router in the kernel will use only one gateway. I assume it 
uses the first one you config and once that disappears the second will 
be used.
	The problem is that eth0 will hardly every fail completely. The only 
way it would is if it is directly connected to say a DSL device and the 
device became unplugged. That would completely drop eth0 and then the 
failover would happen. If only the connection upstream from the DSL 
device went down then the Linux box would happily send data to the up 
ether interface of the DSL, router, switch, etc.
	It would also work if you're terminating connections directly into the 
Linux box... using a ds-1 card or something like that. Again this 
assumes that the connections fail totally. A ds-1 usually will, but in 
offices you're sometimes handed ether which terminates on a switch. That 
switch will be up, but the router or upstream connection has failed.
	I can't find much on the /proc/sys/net/ipv4/route/gc_timeout setting, 
but what I did find suggests that it only provides timing as to when the 
route is dropped. It doesn't add anyway to monitor the connection.

It's an interesting little hack assuming I'm right about what he's doing 
with it, but I wouldn't go around implementing it on most networks.

kashani
-- 
gentoo-server@g.o mailing list