Re: A simple question
2003-04-20 07:01:34
Hi,
Daniel Senie wrote:
Example of usage: I use RFC 1918 space for private networks within
server farms. There's no connection to the public network from this
private segment. Servers connect to both public and private networks.
Backups, NFS mounts, SNMP and SQL data flows over the private network.
The isolated network segment keeps traffic off the public network for
performance reasons.
This has become an acceptable use of RFC 1918 addresses. On the other
hand, there is no need to support this sort of use in the v6
architecture. The address space is simply too vast for people to have
to economize like this.
All the world of private addressing is not NAT, regardless of how many
times people say it is.
My major concern that does drive NAT is the expressed desire on the IPng
list to isolate sites from having to renumber. That use of SLs requires
NAT or proxy. Rob may call it crazy, but the demand is there right now.
I am saddened by the fact that Tony's simple question could not be
addressed. Site local addressing in IPv6 is a concept which has been
mentioned in RFC 1884, 2373 and 3513, the progression of Proposed
Standards. This is a string of documents dating back to 1995. For eight
years this concept was apparently considered a good thing. The
discussion on the mailing lists makes it sound like site-local
addressing is a new idea. I'd like to know why it's taken eight years
for folks to decide it's bad. Is it that folks are just now implementing
IPv6? Is it because the documents these eight years never made the
concept clear, and now it appears too hard to implement? In all those
years, has no vendor implemented site local? Are there any other
features we should reconsider as long as we're ripping the documents open?
Quite simply it is something we seem to have forgotten about: learning
from operational experience. Since 1995 the use of private address
space has changed, as has been said. Also, I don't think people fully
understood the ramifications of SLs. I certainly didn't, initially. We
also learned that using EUI-64s as end identifiers has certain privacy
considerations, so this is not a first. I think we do have to take some
care when making changes to the addressing architecture. If the
decision of the group holds, it seems to me that those of us who voted
"YES" have the burden to now provide an alternative that addresses the
real requirements that SLs attempt to tackle.
It is not unprecedented to change or remove a feature as a document
advances through the standards track. Such changes, however, can have
significant impact on already-implemented and deployed solutions. Such
matters should be considered carefully in that light. Perhaps removal of
features should receive substantially more scrutiny after publication on
the standards track.
I agree. The vote was for deprecation. That means we need to deprecate
in favor of something better.
Happy holidays,
Eliot
<Prev in Thread] |
Current Thread |
[Next in Thread>
|
- Re: Architectural Considerations section in specs, (continued)
- policy domains, S Woodside
- Re: policy domains, Stephen Sprunk
- Re: policy domains, Eliot Lear
- Re: Architectural Considerations section in specs, Keith Moore
- Re: Architectural Considerations section in specs, Zefram
- Re: A simple question,
Eliot Lear <=
- Re: A simple question, Robert Elz
- Re: A simple question, Keith Moore
- Re: A simple question, Robert Elz
- Re: A simple question, Keith Moore
- Re: A simple question, Vernon Schryver
- Re: A simple question, Margaret Wasserman
- Re: A simple question, David Conrad
- Re: A simple question, Bill Manning
- IPv6 address space shortages (was: Re: A simple question), John C Klensin
- Re: IPv6 address space shortages (was: Re: A simple question), Dean Anderson
|
|
|