[A2k] We need a new formulation of end-to-end analysis

Marcin Cieslak saper at saper.info
Fri Jan 25 01:38:36 PST 2013

On Thu, 24 Jan 2013, Seth Johnson wrote:

> E2E is intuitive: if you want to be able to do things flexibly, those
> things aren't going to be able to get through unless the network of
> networks (one network is another thing, and isn't internet anyway)
> transfers information without regard for what it's doing.

Internet is more "network of networks" than pure end-to-end.
It might have been more end-to-end in the era of hosts (using
RFC 1122 parlance), which were larger multiuser machines
that let the end users (humans) to use the Internet. My first
machine I could access the net c. 1992 was VM/SP mainframe
connected to the EARN/BITNET, later also various Unix
"hosts". When talking about net policies of those time,
DNS and address space providers where much more involved
into well-being of the internetwork. I still remember
my DNS zones being checked by the upstream before
they allowed those zones to be delegated.

The concept of "host" needs to be analysed further,
how it evolved - into millions of PCs and
other end-user devices on one hand, and how "networks"
evolved into huge administrative domains of hosting
companies or access providers (ISPs). This has
created a more hierarchical than traditional peer-to-peer
approach, which currently survives on the BGP (interdomain
routing) level to some extent. That's the price
for massive rollout of residential broadband and
wireless access services.

There are two directions one - more strengthening
of the hierachy - mechanisms like RPKI come to my 
mind and also the move from the "home" NAT to
large scale NATs as we have no IPv4 addresses anymore.

On the other hand, rapid deployment of IPv6 could
help us to move into a more end-to-end, cloud-free
environment with every device to be on its own
on the network. Of course, address and route aggregation
issues will bring elements of hierarchy even here.
And firewall-style perimeter security has become
mostly an illusion already.

And finally, when speaking about humans we should
not forget bots and treat them as the users of
the network as well. This brings some interesting
issues onto the table.

Marcin Cieślak

More information about the A2k mailing list