North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: ARIN Policy on IP-based Web Hosting (fwd)

  • From: smd
  • Date: Thu Aug 31 14:44:09 2000

["Uh oh, Sean must be bored" sotto voce from the crowd...
 "But really, it's network-related!"]

Richard Shetron writes:

| On MS-DOS/Windows, its the last part of the dotted filename, ie junk.txt
| would be a text file of junk.c would be a C cource file.
| Most Unix's are the same.  The extension is the last component in a
| period separated name.

Hmmm...  Funny that someone with a login of "multics" would say that!

: sean (es) ; ln -s /bin/echo foo.c
: sean (es) ; ./foo.c hello there
hello there
: sean (es) ; file * | egrep '(English|International language) text' | wc -l
: sean (es) ; ls *.txt | wc -l
: sean (es) ; uname -sr
NetBSD 1.5E
: sean (es) ; echo 'main() { exit(0); }' > x
: sean (es) ; cc -o y.c -x cpp-output x
: sean (es) ; file x y.c
x:   ASCII text
y.c: ELF 32-bit LSB executable, Intel 80386, version 1, dynamically linked (uses shared libs), not stripped
: sean (es) ; cc -v
Using builtin specs.
gcc version egcs-2.91.66 19990314 (egcs-1.1.2 release)

So, making this network-relevant, you have to understand what abstraction
you are looking at, and when you are overloading an abstraction - in classical
IP and its useless and too-costly-to-deploy Official Successor, the IP
address serves both as a topological identifier ("where") and an endpoint
identifier ("what").  This is broken.

In "most UNIXes", a pathname ("/bin/sh") serves as a topological identifier,
since it _locates_ the file.   This locator function is overloaded with
a hint about what is at the location, but they are separate things,
because of links (one what has many locations) and because there is 
no rule about (for example) something that has a locator like ".../foo.c"
HAVING to contain a C program file.

Several operating systems try to avoid the overloading of where and what.
Apple's MacOS is a familiar example - the "what" part is not part of the 
folder locator ("where").   

This is valuable since one does not have to _infer_ the content based
on location, nor try to abuse the topology of the filesystem to align
the two.

IPv6's architecture is fundamentally broken because it DOES NOT separate
"what" from "where", inviting the same overloading that leads us into
these discussions about allocation policies and concerns about the
number of progressively-longer prefixes being carried in the global
routing system.

This is, I believe, the point that David Conrad was trying to make.