[9861] in www-talk@info.cern.ch
Re: Improve the traffic condition on the Internet (fwd)
daemon@ATHENA.MIT.EDU (Benjamin Franz)
Thu Mar 21 12:04:33 1996
Date: Thu, 21 Mar 1996 09:10:27 -0800 (PST)
From: Benjamin Franz <snowhare@netimages.com>
To: "William F. Hammond" <hammond@csc.albany.edu>
Cc: megazone@livingston.com, www-talk@w3.org
In-Reply-To: <199603211606.LAA27155@phoebe.math.albany.edu>
Resent-From: www-talk@w3.org
On Thu, 21 Mar 1996, William F. Hammond wrote:
> It was just said:
>
> > . . . Monitoring even small sites shows that nearly all users still
> > download the graphics and text is *not* the majority. . . .
>
> If the "monitoring" is just counting bytes, then that is probably true.
>
> But I suspect that most of the useful information delivered through
> http is either text/plain or text/html. Unfortunately, it's common
> for 10 KB of useful text/html to travel with 30 KB of IMG baggage.
Regardless of whether you consider it _*useful*_ (such a loaded word that
is...), it remains true that the heavy majority of the traffic travelling
to the user *remains* mostly graphics and other pre-compressed material
that will not benefit significantly from more compression. If a *user*
wants to speed up their *local* link - they can turn off graphics in
every browser known to man. This instantly achieves a nearly four fold
increase in browse speed - with no software changes at all.
The original message spoke of *100 bytes per second transfers* being
caused by not the *local* link, but network congestion. When it was
pointed out that nearly all the traffic there was *already* compressed,
the shift was made to the local link - where, ironically, compression is
*even better* due to compression by modems of text *already*.
So whether you look locally at the modem links, or globally at the
backbones - additional compression of text simply is not going to achieve
significant B/W savings in practice. The local links *already* compress
the text, and it is provable that on the backbones you could not achieve
more than a 20% savings by compressing the text - a savings that would be
wiped out by the general network growth within three months.
IOW: It is a lot of work for very little actual benefit. Much better
cost-benefit would be achieved if ISPs would install caching proxy servers
for their users and formed 'mutual support' groups of heirarchial
proxies to improve the proxies' hit rates. A 'mom and pop' ISP may not
see a lot of local hits on their proxy, but 10-20 'mom and pops'
working together could get a big improvement by using a common proxy.
Then take an association of 10-20 'mom and pop' ISPs and form a 'super
association' caching proxy for 10-20 associations....
--
Benjamin Franz
