Ask IBM about VM65104 when you call.

This message may contain confidential and/or privileged information. If you are 
not the addressee or authorized to receive this for the addressee, you must not 
use, copy, disclose, or take any action based on this message or any 
information herein. If you have received this message in error, please advise 
the sender immediately by reply e-mail and delete this message. Thank you for 
your cooperation.


-----Original Message-----
From: Linux on 390 Port [mailto:[email protected]] On Behalf Of David 
Boyes
Sent: Thursday, September 20, 2012 11:12 AM
To: [email protected]
Subject: Re: [LINUX-390] Question on z/VM VMRELOCATION of SLES 11 SP1 guest

Can't test this at the moment, but do the system logs tell you if the 
relocation generates a disconnect/reconnect event for the interface in 
question? IMHO, if it doesn't, then it should, since you're doing the 
equivalent of unplugging the machine, moving it, and then reconnecting it. Even 
if it's reconnecting to the same VLAN, the initial ARP to make sure you own the 
associated IP address should happen to avoid a race where a duplicate IP 
address could come online during the migration. 

Try running arping on the Linux machine via the 3270 console after the 
migration. If that fixes the problem (and it probably will), configure your 
Linux guests to generate a gratuitous ARP when the network interface state 
changes. That should force the FW to forget the old ARP entry and pickup the 
new one. Why the gratuitous ARP is no longer the default is still a mystery (at 
least to me, anyway) -- I can't conceive of a case where this would be the 
wrong behavior, and a ARP flood is not likely to be caused by a loose connector 
in this scenario (all virtual, all the time).

There is probably a good argument that the VM migration code could/should force 
a proxy ARP for a guest when it gets migrated, though. It has enough 
information to do it (the old and new virtual MACs and the associated real OSA 
MACs) and that would at least force other devices to reevaluate any caches that 
might be present, even if the result ends up with the original configuration. 

Worth a call to the support center to discuss, anyway. 

Reply via email to