🚀 go-pugleaf

It is not arbitrarily multiplying by five to spread the results.  It's a
little arbitrary, but the goal is to simulate a typical page on the web. 
We've looked at hundreds of samples, and a page made up of FIVE files, each
averaging about 10K, is as close as we can come to a "typical" web page. 
Obviously there is a broad range here, with some pages consisting of 18 or
20 files and several hundred KB of data total.  But everyone seems to
accept five files and 50KB total as a reasonably good wag at a typical web
page.

This is simply an effort to put test measurements in meaningful terms. 
packet latency in ms and even percentage packet loss is not going to be
terribly meaningful to most.  But that it takes 12 seconds to download a
web page is VERY familiar territory to a lot of people using the Internet. 
And those designing web pages are often quite sensitive to how long it
takes to download the page.  Please don't use OUR web page as an example of
how to do or not do this.  It is admittedly a horror to download.

Jack RIckard

----------
> From: Marc Slemko <marcs@znep.com>
> To: inet-access@earth.com
> Subject: Re: savvis communications
> Date: Friday, November 21, 1997 9:26 AM
> 
> On 21 Nov 1997, Sean Doran wrote:
> 
> > If the data is in question, try reproducing the experiment.
> 
> I'm afraid that is impossible because I don't know what they are
measuring
> and I don't have dozens of sites around the country to test from. 
> 
> What I am thinking of doing, however, is a more basic test, taking the
top
> 5 (or some other number) and bottom 5 from the report and attempting to
> see if, from my location, I can distinguish which group is which based on
> the top and which on the bottom.  Then I can make this available to
people
> and have them see, for their location, if the Keynote results appear to
be
> valid or not.
> 
> > > Oh, did you know that the Internet has a 40kbit "speed limit"?  You
must
> > > not be reading Keynote's press releases.
> > 
> > At some times of the weekday day, on a California-to-
> > Sweden path where the bandwidth choke point is the FDDI
> > rings connecting the workstations to the routers on either
> > end, if you push more than 200kBytes/second using modern
> > TCP implementations, 4k MTUs and MSSes, RFC 1323 windows
> > and timestamps, you are having a really good day.
> 
> It is certainly true that there are paths with better and some paths with
> horrible bandwidth.  Hey, going to certain sites through Sprint has
> resulted in 50% packet loss within Sprint or at the sprint/MCI border for
> months and they regard that as normal operation.
> 
> However, Keynote is attempting to use that number to say that any
> connection faster than 56k is a waste of time right now.  That has zero
> basis in reality.  For viewing most web sites, I can saturate a 56k
modem.
> I can saturate a 128k ISDN connection just as easily.
> 
> [...]
> > 
> > This kind of "Consumer Report" statistic, while perhaps
> > not a diagnostic tool in itself, does underline the
> > failure of many networks to recognize and fix serious
> > routing and forwarding pathologies which are detrimental
> > to decent end-to-end goodput.
> 
> Certainly.  However the design of the tests is for numbers that look good
> instead of numbers which have meaning (what other reason could there be
> for arbitrarily multiplying by 5 to spread the results?).
>