In-Reply-To: 
Message-ID: 
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Status: O
X-Status: 


I led the first project to put Dow Jones on the internet.


	Rex Ballard
	Standard & Poor's/McGraw-Hill
	Opinions expressed do not necessarily reflect
	the Management of the McGraw-Hill Companies.


On Mon, 8 May 1995, Doug Burnett wrote:

> How does a paper go about deciding how big of a server and Internet
> connection they need for web publishing?  I'm sure the starting places is
The rule of thumb is that the usership will grow at about 20%/month until 
the response times get so slow that your customers start leaving.

For our project, we started with a Sun SparcServer 20.  We were doing 
good until we got to about 20 hits/minute.  That was about 4 months ago.

As we (now they) "turn on the meter", circulation will drop, then climb 
again.  The WAIS engine behind the "pretty face" is extensible to support 
multiple smaller databases or striped raid arrays.  The latency is in 
disk and compare routines.  Searching through a database of 2 gigabytes 
takes a few seconds, especially if your in line behind 20 others.

This is why we paid WAIS to give us the "High Octane" server.

> how many access they expect but how do papers come up with that number? 
A really "hot" page takes about 45,000 hits/day.  That's about 90 
hits/minute during "Peak".  You can figure on about 10% of that to start.

> Take a percentage of circulation?
Paper circulation has little to do with wire.  Mother Jones might be a 
relatively small pub, but on the web it's very popular.  MecklerWeb
doesn't even exist as a paper and it's rapidly becoming a leader in 
OL publishing.

> Just guess?
Look at your product.  Find a related newsgroup for each section.  How 
many postings are there?  The proportions are similar to what you will 
find on the web.

> Can some of you more
> experienced web publishers shed some light on this for me?

We actually had no clue how many people would register.  We have had over 
25,000 visitors, and about 6000 regulars.

> What numbers
> did you use in your web publishing projects? 

I've been making my size reccomendations all along.  Get a Pentium PC, 
Linux, and the fastest connection you can get.  Use the free httpd and 
a low-end database to get started, cost - under $2000.

  When you start making money, you can
plan on quickly upgrading to multiple pentiums, NetScape server, and
very fast RAID disk drives, and commercial text or SQL database for more
money - $100,000 to start, $250,000 for a Pyramid with tweaked Oracle 
database, and 200 transactions/second capacity across a single DS3 link.

> Doug Burnett

Rex


From rballard@cnj.digex.net Tue May  9 14:27:50 1995