Conference smurf::ase

1931.0. "redirect ARP requires router ?" by UTOPIE::BRAUN_R () Mon Mar 10 1997 12:44

(cross-posted in SMURF::ASE and TURRIS::DIGITAL_UNIX)
 
Hi all,

my customer had a problem with ARP-cache and ASE 1.3.
In the meantime it has been solved, nevertheless I would
like to get some more info:


2 Digital-UNIX-clusters with 2 nodes in each cluster
(and 2 LAN-interfaces per system)
ASE 1.3, UNIX 3.2

All nodes (and the UCX-testclient, see below) in the same
10.132.0.0 (255.255.240.0) subnet.

After doing a "relocate service" (moving the virtual IP-address from
one node to the other, but changing the MAC-address to the hardware-MAC-
address of the new active system, requiring a redirect_arp), there was
a different behaviour between the 2 clusters:

One common test-client (UCX-system) realized this change only for one
cluster, for the other cluster the old, wrong ARP-cache entry remained
the same, consequently causing connection problems, only solved
by waiting the ARP cache timeout (20 min) or a manual UCX SET NOARP
(deleting the ARP-cache entry)

The customer then compared both clusters: the one not working had
"router=no" in the /etc/rc.config-file.
After doing a iprsetup -s (enable IP forwarding) everything worked
o.k.

Questions:   

   - all concerned IP-addresses are in the same subnet, so there
     should be no routing issues involved !?!?

   - if it's not a bug, but a feature: where is it documented ?


Any help will be highly appreciated.

Thanks,
Ralph