TCP/IP - proposals, research and discussion


Robin Whittle rw@firstpr.com.au 2007-06-18
 

Ivip - a potential new routing and addressing architecture for the Internet

ivip/

This is a new proposal which largely supersedes the SRAM-based forwarding proposal mentioned below.

Ivip (Internet Vastly Improved Plumbing, or any other set of words you care for) is intended to allow millions to billions of independent, portable, IP addresses and subnets to be used by millions to billions of end-users, with the hosts for these being located in any edge-network (provided the edge network has an Egress Tunnel Router to serve the host).

In short, it is intended to relieve the BGP routing system of the tasks of portability, multihoming and traffic engineering for finer-grainedand more numerous address ranges than the 200,000+ advertised prefixes it is currently struggling to cope with.  Unlike LISP, which Ivip is based on, there is no need for each edge network to have its own Ingress Tunnel Routers.  This means that Ivip can be deployed incrementally.


SRAM-based IP Forwarding

sram-ip-forwarding/

I have written an  Internet Draft suggesting that SRAM-based forwarding be standardised on routers for the Internet's Default Free Zone.  I think this is the only way to ensure the hardware of routers can keep up with the increasing number of /24 IPv4 prefixes which are being used, especially since most of this use is contrary to the route aggregation which keeps routing tables and forwarding information bases relatively simple.  A single Static RAM chip stores a separate FEC value for each of the millions of /24 prefixes, so it is not affected at all by route disaggregation.  My proposals cover both IPv4 and IPv6.


Ping probing of IPv4 host density

host-density-per-prefix/

In February and March 2007 I sent random ping packets to every IPv4 prefix, and combined the results with information about what proportion of each prefix was advertised in the global BGP system, meaning these ranges of addresses are operational and connected to the Internet.  The result is, for each prefix 0.0.0./8 to 223.0.0.0/8, a rough indication of the density of hosts (computers, routers etc.) as a percentage of the "advertised" address space.

I wanted to see this, in order to understand how well the IPv4 address space is being used.  Some ranges of addresses are reserved and can never be used for connecting to the Net.  Of the remainder, some is reserved by the IANA and by the Regional Internet Registries.  The rest has been assigned to ISPs and large Internet users who operate Autonomous Systems.  However, not all that space is "advertised".  

I also analysed the distribution of ping-responsive-host densities among samples of advertised prefixes of different lengths.  Graphs, discussion and software can be found at the above page.