[135463] in North American Network Operators' Group
Re: Another v6 question
daemon@ATHENA.MIT.EDU (Owen DeLong)
Tue Jan 25 17:19:31 2011
From: Owen DeLong <owen@delong.com>
In-Reply-To: <AANLkTi=soS+LQoR3JYJxE8N3o9JYoxrL2dvzHZ+e5Geb@mail.gmail.com>
Date: Tue, 25 Jan 2011 14:17:39 -0800
To: Max Pierson <nmaxpierson@gmail.com>
Cc: nanog group <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Jan 25, 2011, at 1:43 PM, Max Pierson wrote:
> Great reply's on and off-list so far.
>=20
> To hit on a few points ...
>=20
> Owen, thank you for catching my terminology blunder there. I =
understand smaller is !=3D shorter. Complete mistake :)
>=20
> Glad to see most have loosened that policy, as I figured it wouldn't =
hold at the time I originally heard it 2 or so years ago.
>=20
> >You really can't map prefix availability to prefix usage.
> >There are 4 billion IPv4 /32s. There aren't 4 billion LIRs that will =
get /32s.
>=20
> I wasn't exactly mapping prefix availability to usage, apologies if it =
came across like that. My crunching did not include host bits. I =
understand I won't see /64's from each of my neighbors down the street. =
(or I shouldn't anyways).=20
>=20
I think you may still be missing my point...
There are way more /48s available than will ever get used.
There are way more /32s available than will ever get used.
You need to look at the actual number of likely unaggregated sites
and the number of ISPs. The sum of those numbers with some multiplier
probably in the 2.5 range is probably about as close as you can get
to an anticipated routing table size.
That will be much smaller than then numbers you get if you crunch
the number of available /48s.
> >As to the IPv4 de-agg, I think that's going to be one of the primary =
causes for
> >an accelerated deprecation of IPv4 once IPv6 starts to become more
> >ubiquitous.
>=20
> I would agree from a SP perspective, however from and end-user and =
enterprise perspective, most don't care about table sizes and will be =
slow to move to v6 until "everyone else is doing it" ... (as in they =
have to now because their facebook status won't be offered over v4 =
anymore). Alot of organizations I've spoken with still don't know what =
v6 even is. I think alot of end-user land and even some enterprises will =
drag their feet on this as long as it costs money for hardware upgrade =
to support v6, also having staff to support v6, legacy v4 apps that =
won't be ported to v6, etc. (I understand there's ways around this, my =
point is the customer doesn't). I think v4 will be around longer than we =
want it to (unfortunately), but time will tell.
>=20
They will start to care when their ISP starts charging them for every =
prefix they inject. If you don't think that IPv4 deagg will lead to an =
IPv4 prefix injection charge once IPv6 is more widely deployed, think =
again.
> To Jusin's point,
> I agree that we will be net-negative for a while. My concerns is =
trying to dual-stack a v4 table and a v6 table. They both will grow over =
time, and until the powers that be "pull the plug" on re-issuing =
returned v4 space (which I completely disagree with), it'll continue =
that way. That's just my opinion though :)
>=20
I don't think IPv4 will continue to grow for all that long. I think the =
plug will get pulled by ISPs desperate to reduce the spiraling costs of =
continuing to support IPv4. When it starts becoming increasingly =
expensive to get ISPs to provide IPv4 services, the rest of the internet =
will begin to move rapidly away from IPv4.
I anticipate this will take about 5-10 years after IPv4 runout at =
ARIN/APNIC/RIPE (which will be nearly simultaneous).
Owen
> Once again, thanks for all on and off list responses!
> Max
>=20
>=20
> On Tue, Jan 25, 2011 at 1:46 PM, Owen DeLong <owen@delong.com> wrote:
>=20
> On Jan 25, 2011, at 10:19 AM, Max Pierson wrote:
>=20
> > Hi List,
> >
> > Sorry to bring up yet ANOTHER v6 question/topic, but this seems to =
be one
> > that I cannot get a solid answer on (and probably won't and in the =
event
> > that I do, it will probably change down the road anyways), but here =
goes.
> >
> >> =46rom the provider perspective, what is the prefix-length that =
most are
> > accepting to be injected into your tables?? 2 or so years ago, I =
read where
> > someone stated that they were told by ATT that they weren't planning =
on
> > accepting anything smaller than a /32. So what if I get my shiny new =
/48
> > from ARIN and am already multi-homed??? Does ATT not want my =
business (which
> > they wouldn't get if the first place, but for argument sake, yes, I =
chose to
> > pick on ATT, sorry if I offended anyone :) I already see /40's =
/48's ,etc
> > in the v6 table, so some folks are allowing /48 and smaller, so what =
is the
> > new /24 in v6?
> >
> Today, the vast majority of providers are accepting /48s in IPv6.
>=20
> Verizon was holding the line at /32 for a while, but, they moved to =
/48
> a few months ago.
>=20
> Let's be clear on terminology. I don't think anyone is allowing =
smaller
> than /48, but, most are allowing /48 and shorter. (shorter prefix =3D
> bigger network).
>=20
> > I only ask due to the fact that ARIN's policy for end-users is /48 =
minimum
> > (which is what i've been telling folks to apply for or applying for =
it on
> > behalf of them).
> >
> That's correct.
>=20
> > Second, as I was crunching a few numbers to get a rough estimate of =
what a
> > global table would look like in say 3 or 5 years after v4 is =
exhausted (I
> > understand that it's completely unpredictable to do this, but =
curiosity
> > killed the cat I guess), and in a few cases, I stopped due to the =
shear size
> > of the amount of prefixes I was coming with. Where i'm getting with =
this is
> > has anyone done any crunching on prefix count for v6 (as in =
estimates of
> > global table usage with the various prefix lengths seen above =
_based_ on the
> > initial allocation of the v6 space (not the entire v6 space =
itself)). I'm
>=20
> You really can't map prefix availability to prefix usage.
>=20
> There are 4 billion IPv4 /32s. There aren't 4 billion LIRs that will =
get /32s.
>=20
> There are 256 trillion IPv4 /48s (roughly). There are not 256 trillion
> end sites that will apply for /48s.
>=20
> The whole point of IPv6 is that the number of prefixes vastly exceeds
> the number of applicants that will use them.
>=20
> To measure the likely content of the IPv6 global table, then, we need
> to look at the number and type of users rather than looking at the
> maximum available number of prefixes.
>=20
> I haven't had trouble reaching anything I care about from my /48
> advertised through Hurricane Electric and Layer 42.
>=20
> > interested to see how long before we have 96Gb's of TCAM/Memory =
(take you
> > vendor of choice) in our routers just to take a full table. (Not to =
mention
> > still having all of the ipv4 de-agg crazyness going on today. =
Seriously, who
> > lets /28 and /32's in their tables today? And this will only get =
worse as v4
> > fades away).
> >
> Yeah, that's not likely to happen. TCAM doesn't scale that way. As to =
the
> IPv4 de-agg, I think that's going to be one of the primary causes for
> an accelerated deprecation of IPv4 once IPv6 starts to become more
> ubiquitous.
>=20
> Owen
>=20
>=20
