Direct "back channel" routing for high performance

  • Thread starter Thread starter kenw
  • Start date Start date
K

kenw

I'm about to install a couple of servers on a 100MB LAN; one as a file
server, the other a terminal server. Both servers have dual ethernet
ports, and can connect directly to each other with 1000baseT (gigabit)
ethernet.

So what I'd like to do is connect a gigabit crossover cable between them,
and have them both use their other ports to connect to the LAN via the main
ethernet switch. That way, the terminal server can access the files on
file server at gigabit speed, rather than being limited to the LAN switch's
100baseT speed. And both can connect to clients via the LAN without
routing through each other.

I've actually hooked this up and made the appropriate (so far as I can see)
routing configuration. I made the crossover cable back channel a separate
subnet, and added static routes so they know to connect to each other that
way. The one thing I haven't done, as yet, is mess with route metrics.

It sorta works. They talk to each other, and I can force inter-server
traffic to take either route (direct or via the LAN switch) by unplugging
one cable or the other. And it makes a major difference in speed.

BUT I can't reliably predict which interface will be used by the terminal
server to communicate with the file server, if I keep both interfaces
connected. Sometimes it's slow, and sometimes it's fast.

The more I think about it, the more I think it's not an IP issue at all,
but actually an SMB/NETBIOS browsing issue. IP routing normally deals with
routes to reach a given IP host address; here, I have a dual-homed host,
and I want my SMB client to use a specific host address. I could use IP
routing to force traffic addressed for either interface to travel via the
high speed interface, and let the file server route internally between its
interfaces, but I'd rather the terminal server's SMB only used the host
address of the file server's high speed port -- unless that channel was
down, of course.

Anybody here ever done this, or seen documentation on how to do it? Seems
like it should be documented somewhere, but I'm not sure where to look, or
what to call it.

/kenw

Ken Wallewein
Calgary, Alberta
(e-mail address removed)
www.kmsi.net
 
Back
Top