[ppml] 2005-1 status
sleibrand at internap.com
Mon Jan 23 19:20:30 EST 2006
On 01/23/06 at 6:00pm -0600, Bill Darte <billd at cait.wustl.edu> wrote:
> From: Scott Leibrand
> > One possible argument (which I'm not sure I completely subscribe to)
> > is that v6 at least has the possibility of multihoming without PI
> > space, using shim6. The way I see it, shim6 will initially be most
> > viable for the smallest sites, the ones that have no chance of getting
> > PI space.
> Scott, The only reason that the smallest sites will have NO possibility of
> getting PI space is if a consensus number of supporters cannot be found to
> support their cause. Why should they be denied and their larger
> counterparts not?
Yes, there's one very important reason: If you give a small site PI space
in the absence of some hierarchical allocation system (like
geo-addressing, which has its own complexities that'd have to be worked
out first), then that PI space will make it into the routing tables of
nearly everyone running BGP. IOW, everyone running BGP will consume
resources to support that small site's ability to multihome. In the case
of the smallest sites, end users, I think everyone agrees that such a
burden on the global routing system would not be justified.
Now I'm not saying that an unrestricted PI space would result in many end
users running BGP, since their ISPs would charge them for the privilege.
However I think the extreme case illustrates the problems introduced by
making PI space the only way to multihome in IPv6.
In contrast, I think that small sites would jump on the chance to
multihome their network without asking their ISP's permission. I believe
shim6 will eventually be available in major OS's IP stacks, and it will
get used. IMO it will be an alternative to running BGP, and a very
attractive one to smaller sites.
> > So with that in mind, I would argue that ARIN's IPv6 PI policy should
> > encourage small multihomed sites to use a non-PI multihoming model
> > (shim6) while preserving the ability for large multihomed sites to get
> > PI space and multihome the traditional way.
> > Is that a consensus position? If not, which aspect of it do you
> > disagree with, and why?
> Scott, I'd say that placing a special burden on small operations, and SHIM6
> from my limited observations seems like a very 'special' burden, to allow
> them to multihome whereas their larger counterparts can avoid the same
> burden with PI...well THAT is what needs to be justified, seems to me.
As I alluded to above, the burden is relative to the number of hosts and
BGP-capable routers you control. If you control multiple BGP-capable
routers and lots of hosts, then shim6 is harder than PI space. If you
control no BGP routers and few hosts, then BGP is more of a burden. I
think the large site / small site distinction is a natural one, not an
> > If so, then we can and should proceed to defining who's big enough to
> > need PI space for sure (regardless of the success of introducing TE
> > into shim6), who's too small to warrant PI space regardless, and what
> > to do about drawing the line in the middle.
> Scott, I'd say it is too early to decide the haves and havenots until we
> understand full the absolute NEED to make the distinction.
I'm not asking us to decide the have nots. I'm saying we should allocate
PI space to users who will have to have PI space regardless. In contrast
to the last proposal submitted at the L.A. ARIN, I think that universe
encompasses a large number of the sites currently running BGP with full
routes. However, I'm pretty sure that universe doesn't *yet* include
anyone not running BGP in IPv4.
> I am in support of enabling v6 an opportunity to succeed, but not at any
> cost. I'd like to know that it is an assignment policy problem
> exclusively before we ascribe an assignment policy solution.
I agree. That's why I think it's essential that we not allocate IPv6 PI
space to just anyone who wants it immediately. I think the evidence is
abundant that large currently-multihomed enterprises who currently run BGP
and have (or qualify for) IPv4 PI space will need IPv6 PI space as well.
The question in my mind is who else has to have PI space. I don't think
that having a /48 from two ISPs should necessarily qualify you, at least
until we give shim6 a chance to target that market.
More information about the ARIN-PPML