Looking at IP Addresses Differently

Sun Nov 9 07:49:08 UTC 1997

It seems that most people look at IP addresses
as hardware-like, host identifiers, that are close
to the "assembly language level" of the Internet,
if there was such a thing.

In an expanded address space, it can help to
look at IP addresses as "opaque handles".
In other words, they are just numbers that
are passed around in the huge distributed
computer system called the IPv4 Internet core.
Low-level routers just happen to agree that
they should advance packets in certain directions
based on the values of those numbers. Routers
are sort of like dumb ALUs that do not contribute
much to the computation being done.

If you look at packets with extended addresses,
it is very much like moving from 8086 real mode
addressing to 386 segmented or 486 virtual
memory addressing. The addresses become
logical offsets and less tied to the hardware
routers. This is especially true, if new address
spaces are arranged around the edges and
the core is simply used for transport.

With this arrangement, the task of the core is to
look at the extended addressing bits and move
the packets from one side of the core to the
other as quickly as possible. The opaque
32 bit handles only become interesting once
the packet gets to a destination where it is
unwrapped for "execution"...using the ALU and
distributed CPU metaphor.

Now, with extended address bits the focus
shifts from the 32 bit handles to the extra bits
and the way they come about. In the IPv8 Plan,
2,048 TLD Authorities are tasked with being
the keeper of the 11 extended address bits.
Each authority can have a different Address
Management plan and the 32 bit handles can
have very different meanings depending on the
agreements between the parties that use these
extended addresses.

The Address Management plan may not have
anything to do with the low-level details of routing,
especially across the IPv4 core. In some cases
there may be very little routing that occurs based
on the 32 bit handle once the packet traverses
the IPv4 core.

In the IPv8 Plan, the assumption is that the IPv4
core provides reliable 32 bit routing and DNS.
The DNS can be used to dynamically discover
which IPv4 addresses should be used as ports
for tunnels from one side of the core to another.
This is handled based on look-ups using the
extended address bits, not the 32 bit handles.
This is why it is useful to tie a TLD to the extended
address bits and why with 11 bits, there can be
2,048 TLDs.

For most of the TLDs, the Address Management
plan or the "addressing mode", using CPU terms,
would naturally be "real mode" (i.e. the 32 bit
handle is indeed a host address and routing is
normal). As people begin to do more research with
true object-oriented platforms distributed on a
large scale, other addressing modes and Address
Management plans can be developed. The DNS
can once again be used to encode the PSW of
the TLD. The Program Status Word can indicate
which addressing mode is being used and other
global information. This allows for variety in the
number and types of modes allowed and the
mode can be controlled globally.

In summary, if one looks at a packet with an
IPv8 43 bit address, they have to look at the
11 extended address bits and they have to map
those to a particular TLD which in turn can be
used to discover the addressing mode in use
which determines how those extended address
bits are used. In most cases, the 11 bits are
used to traverse the core and the 32 bits are
used to route from there, in other cases more
interesting uses can be developed for the 32
bit quanitities which many people view as IP
addresses, but which will eventually become
opaque handles.

Jim Fleming
Unir Corporation
IBC, Tortola, BVI