I slipped up on the calculation. I assumed 2000::/3 was 2000:: to
2fff:ffff:ffff:ffff:ffff:ffff:ffff:ffff. It actually extends to
3fff:ffff:ffff:ffff:ffff:ffff:ffff:ffff. Guess I slipped up on the hex
math. My mind always thought the 6 bone addresses were in the next block,
but of course now that I think of it correctly, they were not, but instead
the final addresses in this block.
So at this rate, Im guessing more like 100+ years before we go on to the
next 1/8 block. After all 7 out of 512 in 20+ years is a long time to go.
In the next 100 years even with EUI64, some of that space could be clawed
back for use if we got rid of SLAAC. Many networks today only keep it
around for android. But as I said earlier, I doubt any efforts will take
place until we use more than 1/2.
Unlike IPv4 with its hard 4 billion limit, in IPv6 there does exist the
possibility of getting more host bits by getting rid of SLAAC and setting
the local part much smaller than the current 64 standard. Each bit
obtained of course doubles the number of available host bits. Even with
the most liberal standard, exhaust is still 100's of years away.
I think route aggregration is going to turn out to be the biggest issue
with IPv6. If the IPv6 blocks become broken up like IPv4, we are going to
have a very hard time with routing tables that grow to billions, trillions
or beyond. This is why I wish the IETF or someone else would come up with
a clean way to do multihoming with PA addresses. Multihoming and PI
addresses are the leading reason for routing table growth.
Albert Erdmann
Network Administrator
Paradise On Line Inc.
On Sun, 5 Jan 2020, Owen DeLong wrote:
On Jan 4, 2020, at 12:41 , hostmas...@uneedus.com wrote:
I understand that there might have been some poor choices made with
IPv6 in regard to address allocation that might lead to a future
exhaust. The main one is the 64 bit network and 64 bit host decision,
considering that it was based on 48 bit ethernet OUI's. I think it
should have been 80 bits of network and 48 bits of host instead. Even
in the largest of networks, 48 bits is clearly overkill. Having the
current /64 is clearly excessive.
This fails utterly to consider the true history of the 128 bit decision.
Prior to the decision to incorporate EUI-64 addressing into the process,
the most likely address size for IPNG (prior to being given version
number 6) was 64-bits. The additional 64 bits were added to the address
fields specifically to accommodate this. At the time, it was believed
that EUI-48 addresses were approaching shortage and that the IEEE was
likely to make future versions of ethernet with EUI-64 identifiers built
into the interfaces. Also at the time FDDI was shipping with EUI-64
baked-in addresses.
The algorithm commonly attributed to IPv6 (placing ff:fe between the OUI
and ESI portions of the address and turning on the locally-generated
(0x02) bit) is actually the IEEE specification for converting an EUI-48
MAC to an EUI-64 address.
Other decisions like giving every node a /48 also add to the greater
possibility of exhaust at some future time. Many players have already
decided to assign less than a /48 to their customers by default.
I can’t see any reason to give every node a /48 and I’ve never seen
anyone do this. I have seen every end-site getting a /48 and I think
that’s perfectly valid. Let us consider that there are roughly 7 billion
people in the world. Given that the number of end sites (physical
structures or tenants in multi-tenant structures) is unlikely to exceed
10x population, let’s allow for a doubling of the population to 14
billion. That would be 140 billion total end sites in the world.
There are 35,184,372,088,832 /48s in 2000::/3. That’s 251+ /48s per end
site for the entire planet in only 1/8th of the total IPv6 address
space.
However, unlike the situation of IPv4, there is still plenty of time to
correct this. Currently only 1/16 of the address space is currently
used for global addresses. When it comes time to assign the next 1/16
of space, we could always tighten up the standards, leading to vastly
more addresses being available per 1/16 block. Adoption of an 80/48
split by existing players would vastly expand their holdings. Also,
adoption of only providing a /48 upon request and defaulting to /56 or
/60 can also vastly expand holdings as well.
Not sure where you got that statistic of 1/16h. There is 1/8th
(2000::/3) specified for IPv6 global unicast addressing currently. Far
less than half of that has been allocated to RIRs so far.
I was talking to JJB of Comcast one day discussing why they don’t
provide /48s to their customers. His excuse was “If we were to do that,
the way our network is structured, we’d need a /12.”
Think about that… Probably the single largest eyeball ISP on the planet
can number their entire customer base in an extremely wasteful manner
(Trust me, there is no legitimate method more wasteful than CMTS when it
comes to address deployment) within a /12.
There are what, maybe a dozen providers that come anywhere near that
size in all the world. Let’s even say there are 100 of them. Guess what…
In 2000::/3, there are 512 /12s, so that still leaves 412 /12s for
everyone that isn’t a customer of a humongous ISP.
We still have plenty of time while only 1/16 of the address space is
being used to address being more conservative in the future.
I doubt we will exhaust the 1/8th of the address space in 2000::/3 in my
lifetime.
Does anyone know what is the utilization rate of 2000::/3 is or where
this data is being tracked?
J> Yes, it is tracked at IANA. I believe two RIRs have received second
/12s for a total of 7 /12s allocated from the 512 in the pool. There are
also some smaller earlier allocations that have been issued to RIRs from
the pool, but IIRC, they total less than a /12 equivalent altogether, so
roughly 8/512 or 1/64th of the address space has been issued to RIRs.
(This is from memory without research, so feel free to look at the IANA
data for verification).
Owen
Albert Erdmann
Network Administrator
Paradise On Line Inc.
On Sat, 4 Jan 2020, Ronald F. Guilmette wrote:
In message <alpine.lrh.2.21.2001031911040....@bigone.uneedus.com>,
hostmas...@uneedus.com wrote:
[IPv6] also brings RIR's
back to their original record keeping role, without having to police the
number of addresses that a member needs.
I am not persuaded that this will be the case. When IPv4 was first
promulgated, I do believe that just about everyone felt that there
was no way in hell that "the Internet" such as it was, or such as it
might become, could ever use up 4 billion addresses. Now admittedly,
things -are- rather different with IPv6, where the number of addreses
is a lot closer to the number of elementary particles in the Universe,
but I do think it is unwise to ever assume that there are any practical
limits on man's ability and/or willingness to waste stuff. In other
words, I think that some amount of thoughtful husbandry of the resource
will always be needed.
Regards,
rfg
_______________________________________________
ARIN-PPML
You are receiving this message because you are subscribed to
the ARIN Public Policy Mailing List (ARIN-PPML@arin.net).
Unsubscribe or manage your mailing list subscription at:
https://lists.arin.net/mailman/listinfo/arin-ppml
Please contact i...@arin.net if you experience any issues.
_______________________________________________
ARIN-PPML
You are receiving this message because you are subscribed to
the ARIN Public Policy Mailing List (ARIN-PPML@arin.net).
Unsubscribe or manage your mailing list subscription at:
https://lists.arin.net/mailman/listinfo/arin-ppml
Please contact i...@arin.net if you experience any issues.
_______________________________________________
ARIN-PPML
You are receiving this message because you are subscribed to
the ARIN Public Policy Mailing List (ARIN-PPML@arin.net).
Unsubscribe or manage your mailing list subscription at:
https://lists.arin.net/mailman/listinfo/arin-ppml
Please contact i...@arin.net if you experience any issues.
