On Wed, 07 Nov 2007 17:13:48 -0800 kashani <[EMAIL PROTECTED]> wrote:
> Buying a single GigE card would appear to be simpler and cheaper > unless you don't have a GigE switch. :-) well, it just so happens I don't. (The 4-port card cost me $0). Some day the file servers will have Gigabit, and so will the switch, and they'll have all the hard drives on a bus that's better than PCI (~ more bandwith). Until then, I want to better support my diskless clients without spending money, and without choking up my network services. >You should not need to do any routing and I'd be surprised if Linux is >actually doing any routing in this case. Of _course_ it's routing. It has to decide what to do with packets in the output queue, doesn't it? Of course if this was all one route the default routing information (one for the subnet, one for the default route) would be enough. What I'm doing here is moving away from the 'one network connection per computer per broadcast domain' philosophy. (Check out the convenient e-book at policyrouting.org if you have interest or are confused. ) That's why this is a routing question. >Additionally when >you initiate a connection from your server it will always originate >from eth0 because 0 comes before 1 IIRC. Just one of those things. It's because of the metric of the routes in the routing table, actually. Without routing, your computer talks to no one. Haven't you ever set up a network connection by hand : ) ? >However >when you initiate a connection from a client to eth1 the server should >respond out the same interface. I'd play around with tcpdump on a >client and see if this is happening like it should be. You'd think so, but it doesn't. Give it a try! You can use a virtual interface and a temporary sshd invocation on it's address to convince yourself pretty easily that traffic is generally routed based on the destination and therefore while incoming traffic will go in the server's specified port, all outgoing traffic will go out the route with the lowest metric that matches the destination ... unless policy routing is configured. Please correct me if i'm wrong. I am not a networking professional, really, but this is how it works for me now, with 2 seperate interfaces. I'm pretty much positive about this. >The machines connect on 10.11.88.0/24 and you avoid any >interface confusion. Only if I refuse to check my work. If I monitor the traffic on the interfaces, I will quickly see that the route with the lowest metric is used for returning traffic. -- [EMAIL PROTECTED] mailing list
