Review - IPNL: A NAT-Extended Internet Architecture

The paper describes IP Next Layer (IPNL), a NAT-extended Internet
protocol architecture. First the paper makes a good point, NAT is one
of the most popular technology/architecture used today, mainly
because it provides address isolation. Nevertheless, one consequence
of NAT is that it breaks end-to-end addressability, this has not been
a big problem thanks that the servers are the ones that must have
public (and therefore, reachable) IP addresses. On the other hand,
this lost of end-to-end addressability has negative effects, it blocks
the introduction of new peer-to-peer technologies and complicates new
protocol and application design. The authors propose extend NAT by
using IPNL to provide end-to-end connectivity and maintaining the
benefits that provides NAT architectures.

According to the authors, the major attributes of IPNL are (1) IPNL
extends NAT, therefore IPNL can reuse the existing NAT
infrastructure. (2) IPNL utilizes Fully Qualified Domain Names (FQDNs)
as an end-to-end host identifier in packets. By Using FQDNs allows
IPNL, to reuse the infrastructure and applications that support
FQDNs.(3) Extends the IP address space such that the globally unique
IP address space forms the high order part of the IPNL address, and
the private IP address space forms its low order part. And (4) IPNL
isolates site addressing from global addressing.

The architecture that the authors propose is quite similar to NAT with
some variations. The main difference is that IPNL requires
modifications on the end hosts, because IPNL adds a new layer in the
IP TCP/IP stack, this layer is between the network and transport
layer. Which may be a problem to deploy the solution. IPNL headers
carry two kinds of routeable addresses one is the FQDN of the host,
and the other is the IPNL address of the host. FQDN works as static
long-term address that must not change during the connection. The FQDN
address would be the preferred address used by applications. The IPNL
address is a dynamic address that can even change during a connection,
the trick is to bind dynamic IPNL address to a FQDN.

The paper gives a good overview of how IPNL would work. Nevertheless,
the solution is on a early development stage, and there are issues to
solve. For instance, all IPNL routers must know about every host in
its attached NATed networks and it also must know about every host in
the NATed networks of other neighbor IPNL routers! This is a huge
requirement that should be solve before consider IPNL as a suitable
solution. Finally, the evaluation provided by the authors of the
prototype its insufficient and useless to the reader, they say only
say: "we could see no degradation in the throughput at
all".
Received on Thu Nov 30 2006 - 11:58:34 EST