On Sun, 12 Aug 2018 09:44:58 -0400, "Mayayana"
wrote:

"chris" wrote

| It's easier to give everyone a unique IP address. Large scale NAT is
| called "Carrier Grade NAT" (CGNAT)

Ah. That's the term. I didn't know that. Looking
around online it looks like it's not so common, but I
didn't find definitive stats.

Right, it's not common because of the potential side effects.
Implementing CGNAT could disrupt the customer networks or the ISP
networks, or both.

From https://en.wikipedia.org/wiki/IPv4_address_exhaustion, with the
most relevant part being the third paragraph:

[quote]
Transition mechanisms

As the IPv4 address pool depletes, some ISPs will not be able to provide
globally routable IPv4 addresses to customers. Nevertheless, customers
are likely to require access to services on the IPv4 Internet. Several
technologies have been developed for providing IPv4 service over an IPv6
access network.

In ISP-level IPv4 NAT, ISPs may implement IPv4 network address
translation within their networks and assign private IPv4 addresses to
customers. This approach may allow customers to keep using existing
hardware. Some estimates for NAT argue that US ISPs have 5-10 times the
number of IPs they need in order to service their existing
customers.[87] This has been successfully implemented in some countries,
e.g., Russia, where many broadband providers use carrier-grade NAT, and
offer publicly routable IPv4 address at an additional cost.

However the allocation of private IPv4 addresses to customers may
conflict with private IP allocations on the customer networks.
Furthermore, some ISPs may have to divide their network into subnets to
allow them to reuse private IPv4 addresses, complicating network
administration. There are also concerns that features of consumer-grade
NAT such as DMZs, STUN, UPnP and application-level gateways might not be
available at the ISP level. ISP-level NAT may result in multiple-level
address translation which is likely to further complicate the use of
technologies such as port forwarding used to run Internet servers within
private networks.
[unquote]


|
| And IP4 addresses have already run out.
|
| No, they haven't.
| https://en.wikipedia.org/wiki/IPv4_address_exhaustion
|

Your link says they have.

It's more complicated than that. Large blocks of IP addresses,
especially classful allocations (Class A, B, or C) are no longer
available. The use of CIDR allows allocation of much smaller blocks, but
even so, what you need might not be available, so what happens is that
other organizations might return part of an earlier allocation, thus
making it available to be allocated to you, or an organization might go
out of business and sell off its address allocation, (for example,
"Microsoft bought 666,624 IPv4 addresses from Nortel's liquidation sale
for 7.5 million dollars in a deal brokered by Addrex)."

Also, a reminder from the quoted section above where they mention that
ISPs typically have 5-10 times as many addresses as they currently need,
to allow for subscriber growth, so "exhaustion" or "depletion" are both
too harsh to describe the current situation, and all of this only
applies to IPv4. IPv6 totally changes the landscape.

[quote]
Reclamation of unused IPv4 space

Before and during the time when classful network design was still used
as allocation model, large blocks of IP addresses were allocated to some
organizations. Since the use of Classless Inter-Domain Routing (CIDR)
the Internet Assigned Numbers Authority (IANA) could potentially reclaim
these ranges and reissue the addresses in smaller blocks. ARIN, RIPE NCC
and APNIC have a transfer policy, such that addresses can get returned,
with the purpose to be reassigned to a specific recipient.However, it
can be expensive in terms of cost and time to renumber a large network,
so these organizations are likely to object, with legal conflicts
possible. However, even if all of these were reclaimed, it would only
result in postponing the date of address exhaustion.

Similarly, IP address blocks have been allocated to entities that no
longer exist and some allocated IP address blocks or large portions of
them have never been used. No strict accounting of IP address
allocations has been undertaken, and it would take a significant amount
of effort to track down which addresses really are unused, as many are
in use only on intranets.

Several organizations have returned large blocks of IP addresses.
Notably, Stanford University relinquished their Class A IP address block
in 2000, making 16 million IP addresses available. Other organizations
that have done so include the United States Department of Defense, BBN
Technologies, and Interop.
[unquote]


I'm saying that a single visitor can have multiple
IPs. Search bots do that routinely.
....
I've also seen cases where remote locations share.
For instance, a visitor from Acme in one location visits
a page, but a distant location downloads the files.

Those cases will almost never be true for human clients. If the goal is
to walk a site, though, then it's certainly possible and even likely
that the tasks of retrieving initial html documents could land on one
server farm for parsing, then embedded links could be assigned to other
farms for further parsing and ultimate retrieval until all page elements
have been accounted for. Everything would get correlated in the end, but
while it's happening you'd see requests from who knows where.

I believe part of what you're seeing is simply GSLB in action.

[quote]
Global Server Load Balancing (GSLB) is a technology which directs
network traffic to a group of data centers in various geographical
locations. Each data center provides similar application services, and
client traffic is directed to the optimal site with the best performance
for each client. GSLB monitors the health and responsiveness of each
site, and like Server Load Balancing, directs traffic to the site with
the best response times.
[unquote]



-chris