Re: ipAddressTable in new, version-neutral RFC2011 draft

[Mike MacFaden <mrm@riverstonenet.com>]

On Mon, Feb 03, 2003 at 12:09:56PM -0500, Kristine Adamson wrote:
>  On our platform we support IP addresses that exist on multiple hosts but
>are only owned (i.e., advertised to routers) by one of these hosts.  They
>exist on the other hosts as part of a workload balancing function for TCP
>connections.  This has caused problems for some network management apps
>that provide topology discovery information using data obtained from the
>SNMP ipAddrTable from RFC2011.  When the management app sees the same IP
>address in the ipAddrTable from 2 different hosts, the app thinks the IP
>address has moved. Is there a recommended way of reporting these addresses?

Hi Kristine, 

Is the mac address is reported for each entry in each agent the
same or different? Do you cable all the hosts on a common layer 2 
subnetwork or not? 

Some suggestions:

If you want to be compatible with existing management applications
then the host that "owns/advertises" the ip address should
have an entry in the ipAddrTable and the others should not. 
Most mgmt applications still think IPv4 addresses are globally
unique and do not handle the semantics of rfc1918 addressing. 
The backup/redundant systems might report the IpAddresses
in an enterprise mib module and then tie to a separate ifEntry
of which you use a propVirtual(53) or similar for stats/etc.

You might want to review RFC 2787/VRRP-MIB. In this case 
two routers can use the same ip address though not at the 
same time. 

> Now we are looking into implementing the ipAddressTable from the
>replacement for RFC2011, draft-ietf-ipv6-rfc2011-update-01.txt.  Would it
>be appropriate for us to use values of either the ipAddressOrigin or the
>ipAddressStatus MIB objects for these IP addresses, to indicate that they
>are not really owned by a host even though they appear in that host's
>ipAddressTable? If so, which values of which MIB object would be
>appropriate for us to use? And should we use a null RowPointer for the
>ipAddressPrefix MIB object for these IP addresses?

I don't think your case is covered semantically this draft so  
you will run into the same problems as IPv4.  

Regards,
Mike MacFaden