Re: Mailing List Last Call for 2822 update internet-draft

Hi Alan,
At 13:48 26-07-2007, Alan Barrett wrote:
>It has happened in the past that software vendors have created
>message-id algorithms that did not guarantee unique message-ids in the
>face of non-unique host names.  The common wisdom has been that it's
>better to generate message-id's of the form <${something derived from
>a random number, a timestamp, and the non-unique hostname}@${something
>that encodes the software version}.${vendor's domain name}> than to
>generate message-id's of the form <${anyhing}@${non-unique hostname}>.

Having the hostname on the left or on the right in terms of 
uniqueness would not be much of a difference as you face the same 
situation when doing a binary comparison of two message-ids.  The use 
of the domain name/IP address on the right is based on the assumption 
that it guarantees that the message-id is globally unique.  That 
doesn't always hold nowadays with NAT.

>The intent behind my suggested wording is to make it more clear to
>software vendors that the domain name on the right hand side of the
>message-id is allowed to be the vendor's domain name; that it doesn't
>have to be the domain name of the host that's running the software, and
>that inability to know the domain name of the host is not necessarily a
>problem.

The rationale of having the domain name on the right is that it 
guarantees uniqueness of the left-hand side within the scope of that 
domain.  If there is a recommendation to use the vendor's domain on 
the right-hand side, it lessens the scope for uniqueness within that 
domain.  If the domain name of the host cannot be determine, an IP 
address could be used.  Your message-id example is good example for 
generating a unique identifier.  You could have the software version 
and vendor's domain name on the left though.

Regards,
-sm