[afrinic-resource-policy-discuss] Re: [resource-policy]AfriNICPolicy Proposal: IPv6 ProviderIndependent (PI)Assignment forEnd-Sites

[Andrew Alston]

Hi MacTim


> Yes, by all means, I'd like to hear from those African LIRs who have some
IPv6
> deployment experience especially.

We've deployed IPv6 actively and in production to at least 3 of our
universities so far (University of Cape Town, Rhodes University and CSIR are
all actively running and utilizing IPv6, as well as some of our more
utilized services, including everything on mirror.ac.za, which is also the
only ipv6 enabled in rotation Mozilla mirror at this point)


> Well of course, if we are going to do it, let's keep it small, but I
> would prefer
> giving /32's (a la proposal in RIPE region) rather than /48 (policy in
> ARIN region).
> Small in v6 is on a different scale than in IPv4 world ;-)

However, is there a point in maintaining the concept of a IPv6 LIR in this
case?  If you have a /32, you're free to onward assign, so long as you keep
your route announcements aggregated, done/done, wouldn't this be simpler?

> This is the "Moore's Law argument", which while perhaps true, would still
> have significant cost implications for most providers.

This may be true, but at the same time there are some important
considerations to be had here as well.  Many ISP's are going to be forced to
upgrade their routers shortly ANYWAY, due to limitations as it is.  Surely
they should be planning for this NOW.  (I point to the fact that once the
IPv4 routing table hits about 240k routes, many routes will potentially hit
problems as it is, forcing an upgrade, sort this problem out when you're
sorting out the other problem!)  

> I'm lost here, why would companies using multiple blocks for the
> "sake" of aggregation?

Company X has an office in Cape Town when he starts, he gets a /48 worth of
P.I Space.
He now opens an office in Nairobi, in order to keep his space aggregated at
his Cape Town office into a single /48, he applies for a second /48, he's
now running multiple prefix's to support his aggregation per site.

> I guess it's a matter of economics, why does company "y" need to
> upgrade routers so companies "a thru m" can get PI space?  This is the
> oft- invoked tragedy of the commons that we face(d) in the v4 world.

Yet at the same time, that's a reality in a business scenario, there are
costs to doing business, and one of those costs is keeping up with ever
changing infrastructure demands.  The internet is evolving, the ever
increasing size of internet pipes alone demands infrastructure upgrades.  It
was only a few years ago that the world was still thinking in terms of
1.5mbit and 45mbit pipes for major corporate, and while large parts of
Africa unfortunately still haven't got past this paradigm, the rest of the
world is now thinking in terms of gig, 10gig and 40gig lines and interfaces.
These things require infrastructure upgrades, surely the infrastructure
carrying the routing tables can be upgraded at the same time!


> which isn't yet "sorted" in v6, but will need to be (id/locator split ?)

How long as the world been hearing about multi-homing in v6 that still isn't
ready?  How long are we going to wait for it?  Its well and good to say
wait, this isn't sorted yet, but it will be, the V4 space has a limited
life, look at the time frames involved.  2012 is not far away, not far at
all, can we really afford to force people who want v6 that is multi-homed
for some mythical solution that has been promised for years and never
delivered on? 

> This type of renumbering is supposed to be a design feature of IPv6.
> Provider lock-in is bad for consumers, IPv6 is supposed to make this
> less of an issue.

Unfortunately without P.I v6 space is there really evidence that it does?
We've had to renumber once already, when our ARIN Assigned LIR block was
returned and we moved to AfriNIC assigned LIR block, believe me, it wasn't
all that easy!

> It's not a service IMHO, it's a design issue of the numbering scheme.

> The IETF folk who designed IPv6 parameters discarded the notion of PI
> to improve aggregation.  It's a fundamental aspect of the numbering
> scheme.  I reckon we ought to tread quite carefully down this path.

True... however the people who designed IPv4 came out with the concept of
classful routing as well... how much of that do we see today?  I think we
need to change the mindset beyond thinking that just because something was
designed in a particular way, does not mean necessarily that the design was
perfect, and didn't have flaws in it.  It's well and good to design IPv6
without P.I, but its flawed to do so without offering viable alternatives
that are accessible *when they are needed* to the problems this creates, and
unfortunately there ARE no viable alternatives to the multi-homing issue at
the moment.  It may come, but its needed today, and it's not there.

> These ISPs can get a sub-allocation from an LIR if they wish.

And handle multi-homing how?

> Correct, policies should be based on what the community needs.  I just
> don't see the
> "need" here, yet.   I think that revenue implications should be kept
> in the back of the mind during these discussions though.

I strongly agree about the revenue comments!

In summary, from having actively deployed IPv6 from an LIR perspective, as
well as having worked for one of our customers prior to that and having
received and worked on a smaller allocation, the above are just some of
conclusions and opinions I've come to.  I *wish* there was broader
discussion on these issues, both on these lists and elsewhere, and I really
appeal to the AfriNIC community, please, get involved in these discussions,
what are your opinions?  What are your problems with the IPv6 P.I?  How do
you view the advantages?  Let's thrash this out and finally get some
resolution on it!

Just my thoughts

Andrew Alston
TENET - Chief Technology Officer