[cgl_discussion] new requirement about ARP cache updating on client machine based on Linux

Eric.Chacron at alcatel.fr Eric.Chacron at alcatel.fr
Tue Aug 12 06:13:59 PDT 2003


Thanks for your reply.
Maybe only the authors of the IP stack in the kernel know why this is not
done  !





"Cress, Andrew R" <andrew.r.cress at intel.com> on 08/12/2003 03:02:30 PM

To:    Eric CHACRON/FR/ALCATEL at ALCATEL, "Peter Badovinatz"
       <tabmowzo at us.ibm.com>
cc:    "OSDL CGL Specs list" <cgl_specs at osdl.org>, "CGL Discussion"
       <cgl_discussion at osdl.org>
Subject:    RE: [cgl_discussion] new requirement about ARP cache updating
       on client machine based on Linux



I agree, it would be more robust if the kernel sent a gratuitous ARP on
a change from ifconfig.  Having the middleware do this also works, but
timing is important here, so it is better if Linux does it.  I haven't
tried this on any 2.6 kernels yet, so I can't comment on whether this
has changed.

Andy

-----Original Message-----
From: Eric.Chacron at alcatel.fr [mailto:Eric.Chacron at alcatel.fr]
Sent: Tuesday, August 12, 2003 8:51 AM
To: Peter Badovinatz
Cc: OSDL CGL Specs list; CGL Discussion
Subject: [cgl_discussion] new requirement about ARP cache updating on
client machine based on Linux


Hi Peter,

This is not a slam !

This is the third new requirement i propose on this list since 2.0 has
been
closed, and i hope having more feed back
from the lists than for the two first.
Note: The two first are: 1) scheduling priority configuration on Linux
internal threads
            2) RR quantum configuration.

This requirement is more cluster related and more precisely HA related.
I have the feed back that currently when one want to migrate an IP
address
from one machine to
another one on a local network using ifconfig, the ARP caches on the
clients remains with
the last value for a while, whereas ther's a need to switchover on the
new
address immediatly.
So , it seems that Linux doesn't send a gratuitous ARP upon ifconfig
like
some other OS ( Solaris, Tru64 ...).
Then HA implementations have to send this message "by hand".

I would like to get a confirmation of this behavior for the last Linux
releases, and ask for a change in Linux if every body agree.

Best regards,
eric



_______________________________________________
cgl_discussion mailing list
cgl_discussion at lists.osdl.org
 http://lists.osdl.org/mailman/listinfo/cgl_discussion








More information about the cgl_discussion mailing list