Stephan Kulow wrote:
Hi,
I updated my scripts to rescan the last 8 weeks of apache logs for 11.2 data
(it took less than 3 days - we're talking 15G of compressed apache that needs
to be evaluated and my poor workstation is doing alot of other stuff :)
I gather by now, a feature request has been open to help with statistics
which does not require such labour ad PC time and cost.
If cannot find the feature request, if it exists, and I un-subscribed
from this list weeks ago.
If we do create a feature request, the only reliable static element to
count may be the MAC address of the PC's cookies that contacts any
automated counter. If we use the MAC address of a NAT'd group of PC's we
only count 1PC when there could be 100 behind this one NAT'd and MAC
address.
To expedite this feature request, if one exists, could be please
consider counting unique MAC Address of Every PC via the PC cookie.
Counting IP addresses is rather useless especially if hundreds of PC's
exist on any LAN that fall into the black hole of 192.168.?.? of NAT'd
single IP's of the Modem/Router.
I can only speak from Australia's point of view, but the only time we
use a public IP Address is for any Server that requires either Public
Access or Internal/Public Access and even them we use a lot of Dynamic
DNS for HTTP Servers as it is so easy. In just about ever other
condition we use NAT. Government, in the past, for audit purposes did
use public unique IP's for the LAN. This however is slowly becoming less
and less, in-favour of NAT'd IP's on the LAN with MAC Identification and
MAC filtering of open DHCP LAN's and WIFI PC's.
Here in Australia the only unique thing we can guarantee is the MAC
address, IP's can be changed and are mostly NAT'ed having hundreds of
PC's behind the Modem/Routers and an unknown number of VPN Linked PC's
behind the same single NAT'd IP.
Statistically speaking, I would recommend thinking alone the lines of
counting PC's MAC address via the PC's cookie as the only true reliable
way of presenting statistics. MAC address are the only possible, non
transportable and unique to every PC in the World. IP Addresses are
potential unless and
misleading and I have not even mentioned DMZ LAN's with possibly
hundreds of PC's having a DMZ address.
I would be grateful if another could add this as a possibility to the
feature request for gathering of statistics which may be more accurate
and less labour intensive.
Many Thanks
Scott
The other distortion of a PC's IP comes from any VPN link, where the VPN
link is behind a single NAT'd IP and there may be hundreds of NAT'd IP's
behind each VPN Link.
http://en.opensuse.org/Statistics now has preliminary
11.2 data (I will reset
them when 11.2 is out, for now the goal is to test the statistics - I found
already smaller problems caused by aria2c now having a longer user agent)
Anyway, what we see is that there are around 1000 new factory installations
per week and that's great - we can assume some do fresh installations into
VMs, but the number still indicates a lot of factory users.
And 11.1 is still growing, which is also good - as 11.2 will still take some
time (and that will also mean that many will not switch to 11.2 as they just
found the time to update to 11.1 :)
But
http://ktown.kde.org/~coolo/repos.png implies that the 11.1 hype is over,
we reached again a limit we had before the 11.1 release of IPs accessing our
servers (every IP counts only counts once for the oldest product). Then again,
the traffic always goes down in summers (both the nothern and the southern
hemisphere summer - some call the second summer christmas, but that is
politically incorrect I guess :)
The week in march when download.o.o was down (and I miss 60 hours of logs),
had only half the IPs, so this is a neat reminder that dialup users make it
impossible to guess how many users there are. We see on average 5 IPs pro
cookie. So there are between 80.000 and 1.000.000 11.1 users - any more
concrete number is only a guess I'm afraid.
http://jspaleta.livejournal.com/42464.html claims their user base is 115%
of the sum of total IPs for fedora. That again would give you 6,5 Million
openSUSE 11.1 users and 9,2 Million users of 11.0 -
And live cds are still a mistery - in 8 weeks there were only 9 live
installations found, but that maybe because people do not do updates
in these live installations. I have no other explanation atm, but that
mistery is the reason I do this 11.2 logs parsing that early.
Greetings, Stephan