22 Jun
2013
22 Jun
'13
10:29 a.m.
On Fri, 21 Jun 2013, Liam Proven wrote:
Those that tamper with the headers more often than not either strip the
useragent or replace it with their own useragent.
No, it isn't unheard of, rather it is actually quite a common practice due
to bonehead server administrators running software such as 'mod_security'
along with rules that only allow browsers with "popular" useragents to
access sites (on the false premise that the other useragents are faked
and/or are "hackers" trying to break into their website, and that
"blocking" them makes their website more secure). Assuming that this is a
rare 1990s practice and that no-one really does this anymore would be
rather foolish. In fact, I ran into this issue with Consumer Reports
website last year and took the time to try to educate them. Their reply?
"We outsource our web administration and they can't fix it."
Moreover, these browser-based usage share meters all share another common
fault -- they only make one report per IP address per sample window, which
is usually 24 hours (although some only take one sample per IP every 7
days). This means if you have a NAT router (who doesn't?) and have say a
Mac, and Windows PC, and a Linux workstation, only the first used in that
sample window is going to be recorded, and there is no real way of knowing
which website you visit is doing the sampling. If you also have a
smartphone connected to your NAT via WiFi, then that smartphone's
useragent is going to be the one recorded for your IP during the sample
window. This affects usage numbers for the "less common" operating systems
more than the "more popular" operating systems, because operating systems
such as Microsoft Windows have a larger market share.
On 21 June 2013 15:21, Tothwolf <tothwolf at
concentric.net> wrote:
We are talking about usage share. Just as you cannot accurately measure
the number of servers out there that make use Linux, you similarly cannot
/accurately/ measure the number of workstations which make use of Linux.
Most servers do not show up in browser and http
server estimates
Who's talking about servers? We all know Linux is really strong in
servers. No argument there. Web browser
and server statistics can not and do not account for the
majority of the computers out there because huge numbers of them are
not being used for web browsing
I submit that most desktops *are*.
Similar for web browsing, since many of the
larger ISPs are using
transparent caching proxies of their own.
If those modified the browser user agent, then they'd probably
/inflate/ user figures. But I don't think that they do. Either way, such things would affect /all/ users on
all platforms, no?
Yes, however given Microsoft's market share, it is going to skew the
numbers for operating systems such as Linux more than it would for
Microsoft Windows.
That isn't
even taking into account that many Linux users end up
configuring their browsers to report false user agent strings
Very very rare, IME.
in order to force the numerous broken websites
out
there to still work with browsers other than Firefox and Internet Explorer
under MS Windows.
That way something the hardcore did in the 1990s. It's unheard-of now.
I see a broken site a few times a /decade./