Slashdot Mirror
87M Hosts on the Internet?
NTT writes "The Telcordia Internet Sizer provides daily updates on the size of the Internet. The Telcordia solution to quantifying Internet growth statistics is based on an internally developed unique sampling method. In this approach, over 150,000 randomly generated IP addresses are sampled on a daily basis and checked for their existence. Check out the other stats they have here"
How do they account for IPs not evenly distributed? Say some countries have used more of their IPs than others.
What about many computers with the same IP, or many IPs pointing to the same computer?
---
There are many instances where internal hosts (that is, those behind a firewall) have real registered address space addresses. RFC 1918 addresses are nice, but not even close to every company uses them, even for their internal network.
Also, this test doesn't really consider network address translated addresses with public DNS entries. For example, suppose I have an address for www.mydomain.com with my own authoratative domain server. The address is, say, 172.16.1.1 and anyone can connect to it. However, I actually have my firewall round robin the requests for that address to my web farm of 10 machines, 192.168.1.1-10, none of which are in external DNS. The survey would only catch one address, which actually has *no* machine directly associated with it. DNS is a reasonable measure of the size of the internet, but it is hardly an authoritative one.
This isn't even counting DMZ machines (those external to firewalls) that are connected to the internet "directly", but don't have a DNS entry. Why would you want a machine like that? Well, how about IP addresses on routers? Would you want those in DNS? How about intrusion detection servers, which monitor incoming traffic for attempted break ins. You really want to make yourself publicly known, making it easier for script kiddies to find you?
A better test would be an aggragate test of DNS reverse resolution, ping & traceroute. I'm sure that there are many machines out there that are open to some of these but not all three.
My UID is the product of 2 primes.
They presume reverse DNS implies IP address usage. This is not correct, of course. There are many machines that don't reverse lookup. Also, there are many IP addresses that reverse lookup and aren't there. The most glaring data is to look at Lucent in their enterprise list Apparently, Lucent has 48 machines for each employee. Lucent will successfully reverse DNS every IP that they are asked about, into something like h135-1-1-1.outland.lucent.com. Splitrock.net apparently has a similar scheme, although the naming method is a little more opaque.
When your estimate is 87 million, of which 8.3 million of your count are highly suspect, it's not the 3 per cent sampling error that you should be concerned about.
Mine just dials my usual selection of ISPs. As it happens one of them is a fixed IP dialup (demon internet since the have mobile phone access numbers for very fast conencting) and what I was pinging was definitely my phone since the ping time was about 900ms from a cable modem and when i hung the phone up the responses stopped.
:( Does anyone else know any particularly Wap friendly 0800 isps in the uk?
Since my freephone ISP (lineone)are stopping access soon i'll have to start paying for access again soon
I'm always curious as to why people are interested in the size of the internet. As long as it works, and people think it's running nicely, does it really matter? I can't see any competitors to the internet for various institutions to be battling down, so I'm assuming that these reports are issued as nothing more than a cheap way to get some hits on reporter's website and to raise their meagre profile a little.
;-)
But then, I've always been cynical like that.
Thoughts?
Now that this survey has shown us how deep the penetration of the Internet is, let's figure out the stats for the real meaning of the net: PORNO!!!!!
Assuming that each host were to have 10 megs of original, non duplicate pr0n online (yes I know thats a very very low estimate), with 87 million hosts out there, that would mean that there 830 terrabytes of luscious luscious pr0n out for your downloading pleasure! Excuse me while I check out the newsgroups...
"The most fortunate of persons is he who has the most means to satisfy his vagaries."
"The most fortunate of persons is he who has the most means to satisfy his vagaries."
- Marquis De Sade
I wonder if they take into account which netblocks have been issued by regional registries like ARIN
With all the mostly unused but allocated Class As and Class Bs that were given out long before we ever knew how popular the net was going to be, firewalls, masquerading, dynamic IPs and God knows what else, how good can sampling be on this network?
--
Care about electronic freedom? Consider donating to the EFF!
This is a good point. If they do a scan at an IP address and none of the priviledged ports are responding (accepting connections or indicating that they're closed), the best you can do is assume that there's no computer there. Right?
How will this have skewed the results?
Fire and Meat. Yummy.
Does this measure all IP's, or just addresses in DNS or what? If its all IPs then that means theres only 87000000 out of a possible 256^4=4294967296 which means we are only using 2% of the possible address space. So why all the noise about IPv6?
If I remember correctly from my stats class the general rule for margin of error is actually 1/sqrt(n). Common sense says that as the sample size increases, the margin of error should decrease, so sqrt(n) doesn't seem right.
1/sqrt(n) gives you the margin of error as a percentage, to figure out the number of IP addresses which this accounts for, the calculation is 1/sqrt(n) * n which becomes sqrt(n) which is what he was talking about.
Anyway, a sample size of 150,000 is incredibly good, and I think margin of error will be so small that it's not worth calculating
It's 0.258%
Hell, that means we can put off IPv6 for another 20 years or so, right? :)
DrLunch.com The site that tells you what's for lunch!
My sister says she has the internet on her iMac, which has only got a 6 gig HD, so how the hell could it be growing if she's never upgraded?
Acting stupid isn't much fun when there's someone around who knows better
sig:
sig:
See the "..for smart people" banners Wired runs here? Look elsewhere guys.
I have a site with a 7 IP block ... only the gateway IP address is accessible by pinging, etc.
Doing a full service scan of them might reveal something, but that would be dangerous considering the number of people who take a dim view of being probed.
- Michael T. Babcock (Yes, I blog)
These behave very strangely. The TCP/IP stack on my nokia 7110 does respond to pings (although i'd hazard a guess that not all models do. It does produce very odd results in nmap - which kicks up dozens of errors about unrecognised responses.
Remember that nokia expect there to be a worldwide market for 500 million wap enabled phones!? that'll eat into the ip space.
Anybody know what the margin for error with this thing is? I mean, with the millions of possibilities for IP addresses, how accurate can a "random sampling" of 150,000 IP addresses?
Just curious, 'cause I couldn't find an exact number that they published on the site...
ipchains -P input DENY
to be contiunued...
long live IP address surveys.
That is besides all poor souls writing lame messages on Slashdot from a MASQed machine.
Baker's Law: Misery no longer loves company. Nowadays it insists on it
http://www.sigsegv.cx/
This only measures how many hosts are listed within DNS, not the total number of machines on the internet. It doesn't measure IPs used by dialups, machines behind firewalls, IP masquaraded machines, etc. In other words, there are more than 87 million computers on the internet, quite a few more I would guess. In fact, I would say that the exact number is almost impossible to figure out.
what if a machine wasn't up when it was checked... could there be a lot of machines that just arn't connected all the time?
150,000 randomly generated IP addresses are sampled on a daily basis and checked for their existence
My firewall hides (NATs) blocks of hundreds of IPs, and no 'check' will confirm their existence or non-existence.
M$: "We're #2!"
From the nmap man page:
:). It will never end. This can be useful for statistical sampling of the Internet to estimate various things.
"-iR This option tells Nmap to generate its own hosts to scan by simply picking random numbers
If you are ever really bored, try
nmap -sS -iR -p 80
to find some web servers to look at."
The only difference is that most normal people aren't bored enough to keep going after the 500th or so 403 Forbidden error.
And 'mid this tumult Kubla heard from far
Ancestral voices prophesying war!