[LINK] IPv4 host density measured by ping

Adam Todd link at todd.inoz.com
Thu Mar 1 00:30:33 AEDT 2007

At 10:32 PM 28/02/2007, Robin Whittle wrote:
>I couldn't find any recent research on how many computers exist on the
>Net, so I did my own:
>   http://www.firstpr.com.au/ip/host-density-per-prefix/
>I estimate about 107 million computers acknowledging pings.  Does anyone
>have insight into computers which are genuinely on the Net but either do
>not acknowledge pings, or have those acknowledgements filtered out by a

Many many many ingress points now reject ICMP ping packets because of the 
volume and potential for ping floods.

You will have failed to detect most of the hosts in my IP address space, as 
I have filtering on all but a select few IP's.  Generally only one IP 
address that is available on a server, but some of my servers are 
"invisible" to ingress traffic, and are visible to egress.

Don't forget, behind some of those IP addresses are hundreds of 
workstations NATING and PROXYING via a single IP :)

I have 8 workstations with live internet access that are on a private 
subnet NAT'ed.  I could at any time choose to give them visible IP addresses.

Actually that wouldn't be so easy as I'm tight on address space on my 
visible subnet in this location!  I might squeeze 3 onto visible IP's.  I 
could drop two in at the border router which has a /28 active on it for 
testing purposes, but that would leave windows horribly exposed and put all 
my protection behind the workstations :)  I prefer to run everything into 
my Firewall, DMZ, Visible Network, DMZ, firewall, private subnet :)

>I surveyed the average host density for BGP advertised prefixes of
>lengths /8 to /24.  There was considerable variation 4:1, not counting
>the atrociously under-used 19 /8 prefixes.  Then I measured a bunch of
>prefixes individually so I could see the distribution of host densities,
>for instance in a sample of /24 prefixes.  About 1/3 of /24s did not
>return a single acknowledgement when I pinged every one of each prefix's
>256 addresses.

Well you won't get much on .0 or .255 as most people filter those.

If the /24 has been masked with then you'll have 1/2 of the 
address space available for hosts, 1/4 for network and 1/4 for 
broadcast.  Very inefficient, I know, but people still do that instead of 
VLSN routing.

I actually have a /29 that feeds the remainder of the /24 internally using 
VLSN, however, you won't be able to ping the other 60 active hosts and the 
transional 80 odd addresses, unless you are inside my network.

>Some IPv4 space can never be used.  Some is still in reserve, and is
>being handed out to RIRs at about one /8 prefix per month.  That will
>probably run out around 2011 or so.  Not all the space allocated to
>users (ISPs and organisations with AS numbers) is currently advertised
>on the global BGP routing system.  Of the space which is advertised,
>many of the advertised subnets have few if any computers connected to them.

Or so it might seem.  There are some IP address hogs around.  I have a 
large subnet allocation of which much is in use, some because of legacy 
software, some dynamically allocated and some that to my own frustration 
can't be used effectively because of advertising routing limitations.  So 
we loose a few sub-subnets, in order to gain a larger address space for a 
project :(

I'd love to be able to advertise /27's :)

>So there is a lot of scope for making better use of already allocated
>space once the supply of unallocated addresses dries up.

Well some scope :)

More information about the Link mailing list