Personally, I'd avoid the 'load balancing' portion of the bonding
driver as I've had all sorts of quirks with some of the 'cheaper'
switches out there.
The active-backup stuff works very well. :)
On 29/08/2008, at 5:12 PM, Michael Alger wrote:
On Thu, Aug 28, 2008 at 06:30:21PM +0200, Raúl Hernández wrote:
I have got a question ip failover when our node is a client in the
network. Some ASCII to get the picture:
Linux box
+----------------------------------------------+
/ NIC1 ---------> switch1
Protocol Logic /
\
\ NIC2 ---------> switch2
+----------------------------------------------+
The thing is that my process ("protocol logic" above) tries to keep a
TCP session with just one server (which is reachable through
switch1 and switch2) [...] Only one NIC is active and I am trying
to keep the session even if for example the link to switch1 is
down, then my first try is to failover the same ip that NIC1 owns
to NIC2 so the connection and the sessions are kept alive (if they
occur within the protocol timeout).
I think that moving an IP through another NIC does not causes
further problem from the TCP "socket()" perspective but the main
issue I find is that I do not know how to failover the IP
*internally (on the same machine)*. Any suggestion to start up ?
It sounds to me like you want to use the bonding driver rather than
doing this in userspace.
http://www.kernel.org/doc/Documentation/networking/bonding.txt
Most probably you want to use mode 1 (active-backup), although if
your switch supports 802.3ad that could be a good choice as well.
I think to get heartbeat do it you'd need to hack one of the IPaddr
resource agents.
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
