[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: LCUP Issue: UUID



Now I understand. Your desire is for a unique identifier in the universal namespace--the (currently) 128 bit namespace known as a UUID/GUID. I believe this poses a rather significant challenge. 

Most algorithms depend on an IEEE 802.1 node address--commonly referred to as a MAC address--as a unique value. In theory, this is spatially unique. In practice, it is not.

IP networks over Ethernet have emerged as the dominant network combination. These networks do not depend on the universal uniqueness of the node address. The node address need only be unique on the
Ethernet router segment. In our 100,000 node network, we currently have at least 30 cases of multiple host instances 'sharing' node addresses, most courtesy of VM-Ware. Similarly, every LPAR on a
mainframe utilizes the same small handful of available network nodes. Highly reconfigurable systems, such as the Unisys ES7000 series, can freely associate up to 96 PCI channels with any of up to 8
independent partitions, each running a different OS.

We regularly acquire Sun machines with quad network cards. By default, all four ports are configured with the same node address. Very often, each of these nodes is connected to a separate IP/Ethernet
segment so the namespace collision is irrelevant. The node address can be overridden in software when necessary.

In a nutshell, I believe the node address component is no longer sufficient to guarantee uniqueness of a uuid.

Mike Liben


-----Original Message-----
From: Kurt D. Zeilenga [mailto:Kurt@xxxxxxxxxxxx] 
Sent: Wednesday, September 04, 2002 4:31 PM
To: Liben, Michael (GTS)
Cc: Liben, Michael (GTS); 'John McMeeking'; Liben, Michael (GTS); ietf-ldup@xxxxxxx; owner-ietf-ldup@xxxxxxxxxxxx; 'steven.legg@xxxxxxxxxxxxx'
Subject: RE: LCUP Issue: UUID

At 01:07 PM 2002-09-04, Liben, Michael (GTS) wrote:
>-----Original Message-----
>From: Kurt D. Zeilenga [mailto:Kurt@xxxxxxxxxxxx] 
>Sent: Wednesday, September 04, 2002 3:54 PM
>To: Liben, Michael (GTS)
>Cc: 'John McMeeking'; Liben, Michael (GTS); ietf-ldup@xxxxxxx; owner-ietf-ldup@xxxxxxxxxxxx; 'steven.legg@xxxxxxxxxxxxx'
>Subject: RE: LCUP Issue: UUID
>
>At 12:25 PM 2002-09-04, Liben, Michael (GTS) wrote:
>>>The UUID only needs to be unique in the DIT in which it has meaning (the context).
>
>>LDAP and X.500 are designed to support a global DIT.
>
>Is that a global DIT or a global namespace?

Global DIT.

>I believe the DIT would be appropriately partitioned into administrative contexts.

The global DIT can be appropriately partitioned into
administrative contexts. 

Bring this back to UUIDs... the universe of an entry
UUIDs is the global DIT.