[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [GLOBAL-V6] Comments on AP Consensus
>For "is this LIR worthy to get an allocation", I don't see it as useful
>- I said this before: as the kind of customers that a LIR has is so
>vastly different, but the IETF rule says "a /48 to each of them",
>I see any criteria based on absolute numbers as pretty useless.
>
>Examples are "AOL" (horrendous amounts of /48s to single-user
>end-sites - bad usage inside the /48, but lots of /48s used) and a
>national research network in a small country, that serves only
>"a hand ful" of universities (few /48s, but good usage inside of
>those /48).
>
>Not solveable by math.
>
>... but maybe the 200-customer-figure will work.
Yes, if everybody wants it to work and is reasonably flexible ;-)
>Wilfried should say something to this, from the view point of a research
>network...
Gert, you shouldn't point fingers at me that explicitely :-)
I'd like to have a week to spare to properly sort out my thoughts, to
put them into proper words, and to read through all the documents and
minutes (which are better than nothing - but still not the real thing).
In reality I've just got a couple of minutes to spare, because we are
actually building a (somewhat biggish) IPv6 network right now (6NET)....
[well, it looks like I am going to devote some more minutes...]
Gert, in general, you are right with your observation that the
recommendation to use /48, whenever there is subnetting involved, kills
the idea of using "number of assignments" as a measure for the size
(read: *implied* importance) of a network.
I actually provided our situation as input to the discussions in
Bangkok: a medium sized national research network with a customer base
of some 100..200, maybe 300 in 2 years time. With a "site" ranging in
size from a single flat to a biggish university (some 70.000 students
plus some 6.000 staff), and everything in between.
Fact is that we do NOT want an end site to apply for an (s)TLA.
Fact is that any magic (read: *arbitrarily* chosen) number is going to
be *useless* (technically speaking), but if this is the price we have to
pay in order to get a new, and *more flexible*, policy in place, let it
be that way.
E.g. I could cope for a while with one /48 for the needs of Vienna
University, including, say ADSL and cable modem connections to the
students' homes and hostels (and managing them as part of the "site").
I might eventually prefer to apply for a /47, but that's a completely
different story.
However, if someone insists on a particular figure, say "200" (or 776
:-), I would go through the trouble of registering some 300 (or 800?)
students as individual beneficiaries (customers) of an IT services that
Vienna University or ACOnet provides to them. Anyone to argue that a
university's student is less "worthy" than a SOHO customer of an ADSL-
or cable modem-based ISP?!
The same thing with the primary and secondary schools in our little
country. We could (technically preferred!) treat a whole school district
as a site with a /48 (==> 9 federal states and districts), or we could
treat each school, usually having more than one sub-net, as a site which
gets a /48. That's slightly more than 200....
Bottom line[1]?
Well, if someone wants us to play silly games, we should be able to
do that.
Bottom line[2]?
200 sites within 2 years is easy to come up with, but difficult to
argue against.
Bottom line[3]?
In 2 years time, if indeed someone has the motivation to do a
review - the RIRs have to find the resources for that exercise.
That was the reason why I argued against a fixed re-application
interval for the allocations, in favour of a recommendation to the
RIRs to do such a review after 2 years if
1) they do see a *technical need* for that, and
2) if they do have the resources for doing it.
[Those in the RIPE region should be able to understand the
background for that recommendation.]
I hope that the Bangkok minutes do document that recommendation.
If they don't, they should be amended.
Bottom line[4]?
We have spent **wwaayy** too much time on that discussion already
(instead of distributiong addresses and building IPv6-based
networks!), and we should get done with it **nnooww**.
Trying to re-open the thing *right now* has the potential to kill
IPv6 (or the RIRs' involvement in the management of that
address space)...
There are many more things I'd like to add, to clarify and to explain,
but that would mean just another night with very little sleep.
Just one more thing, talking about multi-homing: one of the BIG
differences in V6 (compared to V4) is that a TCP session can easily
survive a layer 3 re-route, because the source and destination address
remain the same.
Trying to do redundancy with multiple addresses on both ends has the
potential to get us back to the X.25 ages: the TCP session breaks, some
sort of RESET. Which means that the application has to get involved.
Unless we can come up with some clever abstraction for identifying
source and destination for a TCP (or V/AE2TP *) connection...
*) Virtual/Abstract End-to-End Transport Protocol
Wilfried.
-
- This list (global-v6) is handled by majordomo@lists.apnic.net
