Re: FW: Application for port-number (more tech)

[Fred Baker <fred@cisco.com>]

At 03:18 PM 5/1/2003 -0700, Bill Fenner wrote:
> Given that the port-numbers registry is the replacement for the Assigned
> Numbers RFC, the application seems reasonable given what the existing
> standards say.
I wouldn't agree that the port number registry replaces Assigned Numbers. 
Michelle has umpteen registries for various things (see 
http://www.iana.org/numbers.html), and together they comprise what used to 
be called "Assigned Numbers".

Apple is assuming that the port number registry registers the ASCII name 
and guarantees its uniqueness, so that a DNS SRV record with that character 
string uniquely specifies that service. This is, specifically, something 
that the port number registry does *not* do. The registry registers the 
combination of a port number with a transport protocol (TCP/25 for SMTP, 
for example), and associates it with a *use*. The *use* differs from the 
character string and is not uniquely identified by the character string. 
For example, take the first few lines in the registry:


        Keyword         Decimal    Description                     References
        -------         -------    -----------                     ----------
                          0/tcp    Reserved
                          0/udp    Reserved
        #                          Jon Postel <postel@isi.edu>
        tcpmux            1/tcp    TCP Port Service Multiplexer
        tcpmux            1/udp    TCP Port Service Multiplexer
        #                          Mark Lottor <MKL@nisc.sri.com>
        compressnet       2/tcp    Management Utility
        compressnet       2/udp    Management Utility
        compressnet       3/tcp    Compression Process
        compressnet       3/udp    Compression Process

Notice that "compressnet", a character string, is associated with four 
different protocols and processes and two different process - a compression 
process and a management process. This is true because IANA doesn't 
guarantee the uniqueness of the character string.

The fact that RFC 2782 incorrectly describes the registry, and the fact 
that you think the port number registry, one registry out of many, replaces 
everything that was once in Assigned Numbers, are both scary. That doesn't 
make Apple's statement correct, nor does it mean that Michelle doesn't need 
an appropriate registry to put this thing in, and therefore instructions 
from <somebody> on what the registry should look like.