Re: What's a "normal" ratio of web sites to IP addresses...

[John McCormac]

On 31/03/2022 23:15, Bill Woodcock wrote:

…in a run-of-the-mill web hoster?

This is really a question specifically for folks with web-site-hosting 
businesses.

If you had, say, ten million web site customers, each with their own unique 
domain name, how many IPv4 addresses would you think was a reasonable number to 
host those on?  HTTP name-based virtual-hosting means that you could, 
hypothetically, pile all ten million into a single IP address.  At the other 
end of the spectrum, you could chew up ten million IPv4 addresses, giving a 
unique one to each customer.  Presumably the actual practice lies somewhere 
in-between.  But what ratio do people in that business think is reasonable?  
10:1?  100:1?  1,000:1?



Not exactly in the web hosting side of the business but I do run a 
website to IP survey for the gTLDs, the new gTLDs and some ccTLDs each 
month that covers approximately 248.3 million domain names.


It is a complex question because the use of IP addresses for websites 
has been changing. Some of the IPs with large numbers of websites are 
actually registrar/hoster holding page websites, sales or Pay Per Click 
parking, DDoS protection, redirectors or load balancers. There is also 
the dedicated versus shared hosting issue which sees large numbers of 
websites on shared hosting and fewer on dedicated hosting with single 
IPs. Virtual hosting also complicates things because it is not unusal to 
see multiple domain names in different TLDs (eg: .COM and .ccTLD) 
pointing to the same IP.


There were 12,300,576 distinct IP addresses (IPv4) in the March 2022 
survey. That also included a small number of private IPs, bogons and 
non-routed IPs.


These are the counts for the top 20 IPs.
34.102.136.180  26308511
3.33.152.1478897990
15.197.142.173  8896940
34.117.168.233  5920870
198.185.159.144 3614480
198.185.159.145 3601589
198.49.23.144   3600433
198.49.23.145   3600334
198.54.117.212  3143453
198.54.117.215  3143451
198.54.117.218  3143448
198.54.117.211  3143447
198.54.117.216  3143446
198.54.117.210  3143445
198.54.117.217  3143444
34.98.99.30 2929772
188.114.97.72708015
188.114.96.72708013
23.227.38.742535730
35.186.238.101  2152424

Some of those are load balancers/redirectors/holding/sales/PPC/DDoS 
protection IPs.


The number of IPs with a single website was 6,943,207. The average 
number of sites per IP was 24.3414. The limitations are that despite the 
large number of domain names in the survey, it is not a complete survey 
of all TLDs (some ccTLDs are not covered). Even though websites may have 
IPs, that does not necessarily mean that there is an webserver running 
on the IP. (That's getting into Web Usage measurement which determines 
how websites are being used or not used.)


Regards...jmcc
--
**
John McCormac  *  e-mail: j...@hosterstats.com
MC2*  web: http://www.hosterstats.com/
22 Viewmount   *  Domain Registrations Statistics
Waterford  *  Domnomics - the business of domain names
Ireland*  https://amzn.to/2OPtEIO
IE *  Skype: hosterstats.com
**

--
This email has been checked for viruses by AVG.
https://www.avg.com

Re: What's a "normal" ratio of web sites to IP addresses...

[Owen DeLong via NANOG]

> On Mar 31, 2022, at 16:47 , Bill Woodcock  wrote:
> 
> 
> 
>> On Apr 1, 2022, at 12:15 AM, Bill Woodcock  wrote:
>> …in a run-of-the-mill web hoster?
>> I’m happy to take private replies and summarize/anonymize back to the list, 
>> if people prefer.
> 
> I asked the same question on Twitter, and got quite a lot of answers in both 
> places pretty quickly.  Thus far, 23 answers, with an average of about 
> 490,000 and a median of 1,500.
> 
> Obviously there are a lot of different factors that go into this, but the two 
> that were cited most frequently were that user who want their own individual 
> IP drive the number down, while large load-balancing/caching infrastructures 
> drive the number up.
> 
> Thank you all very much.  I appreciate the education, and I hope it’s useful 
> to others as well!
> 
>-Bill
> 

I would think that the 490,000 is more likely to reflect “web servers” per 
address vs. “web sites” per address.

I think that your mention of load-balancing and caching somewhat prove (or at 
least support) my speculation here.

I suspect that when you talk about “web sites” instead of “web servers”, the 
number probably falls somewhere in the sub-1k range.

For clarity, “https://www.amazon.com/[ …]” is a web 
site. It is almost certainly served by many many servers.

Prior to SNI, it was mostly 1 web server per address. In 2018, major CDNs were 
just starting to consider
ending support for non-SNI clients.

Owen

Re: What's a "normal" ratio of web sites to IP addresses...

[Bill Woodcock]

> On Apr 1, 2022, at 12:15 AM, Bill Woodcock  wrote:
> …in a run-of-the-mill web hoster?
> I’m happy to take private replies and summarize/anonymize back to the list, 
> if people prefer.

I asked the same question on Twitter, and got quite a lot of answers in both 
places pretty quickly.  Thus far, 23 answers, with an average of about 490,000 
and a median of 1,500.

Obviously there are a lot of different factors that go into this, but the two 
that were cited most frequently were that user who want their own individual IP 
drive the number down, while large load-balancing/caching infrastructures drive 
the number up.

Thank you all very much.  I appreciate the education, and I hope it’s useful to 
others as well!

-Bill



signature.asc
Description: Message signed with OpenPGP

Re: What's a "normal" ratio of web sites to IP addresses...

[David Hubbard]

I don't know that there is a normal as it likely depends heavily on the revenue 
per customer and the service's tolerance for giving out IP addresses.  It also 
depends heavily on the back end infrastructhre and what kind of service is 
being provided.  There's probably massive scale behind Cloudflare IP addresses. 
 There are middleware-style ecommerce and blog platforms where there is the 
same, i.e. lots of sites behind any given IP because every customer receives 
the same service from the same software; likely thousands or more per IP in 
that case.  As you get more custom, probably far less per IP as that's when 
sites tend to start being mapped to dedicated virtual machines / servers, 
shared hosting, etc. where it goes anywhere from a few hundred to one site on a 
dedicated server.

Sorry to go off on a tangent but this got me wanting to rant. __

Still, to this day, SEO "experts" continue to guide clients towards service 
platforms (hosting, ecommerce, blogs, etc.) where they know it remains possible 
to get an exclusive IP address because they are "sure" that will produce 
meaningful search positioning gains.  I started a thread on this topic on nanog 
about this back in what I think was 2003 because every business entity had an 
SEO expert insisting their various websites receive IP addresses on subnets 
that differed enough to be "distant" from one another because Google would 
otherwise penalize them.  I expressed frustration at that because it ensured 
sites that had no technical need for an exclusive IP address would get one 
anyway, wasting a rapidly depleting resource, and costing the provider in the 
process while they could still get address space.

A Google Director, Craig Silverstein, said this wasn't the case, but just 
casually in a slashdot interview.

Matt Cutts later refuted it directly in 2006:  
https://www.mattcutts.com/blog/myth-busting-virtual-hosts-vs-dedicated-ip-addresses/

And he made the point once more in a 2013 Youtube video.

Three semi-official statements on the subject, the most recent nine years ago.  
So, it hasn't done much to dissuade the SEO experts of continuing to steer 
their clients towards places they think an exclusive IP will be issued.  
Fortunately the huge rise of CDN's seems to be getting things back on track, 
because those can produce more meaningful SEO benefit from the faster transit 
to eyeballs, putting exclusive IP recommendations on the back burner.

David



On 3/31/22, 6:19 PM, "NANOG on behalf of Bill Woodcock" 
 wrote:

…in a run-of-the-mill web hoster?

This is really a question specifically for folks with web-site-hosting 
businesses.

If you had, say, ten million web site customers, each with their own unique 
domain name, how many IPv4 addresses would you think was a reasonable number to 
host those on?  HTTP name-based virtual-hosting means that you could, 
hypothetically, pile all ten million into a single IP address.  At the other 
end of the spectrum, you could chew up ten million IPv4 addresses, giving a 
unique one to each customer.  Presumably the actual practice lies somewhere 
in-between.  But what ratio do people in that business think is reasonable?  
10:1?  100:1?  1,000:1?

I’m happy to take private replies and summarize/anonymize back to the list, 
if people prefer.

Thanks!

-Bill