At 10:18 PM 1/23/95 GMT, George Stachnik wrote:
>
>In that context, what would *you* like us to discuss on the
>next TCU broadcast?
>
>-geo
 
  Didn't I here you mention that DCE will be covered on April 19th?
 
  Anyway, George, I was a disappointed by the client/server presentation in
the area of public network support (i.e. the internet, campus networks
connected to the internet, etc.), but not surprised.  I can't believe no
mention (maybe I missed it?) of ported software on jazz.external.hp.com,
such as NCSA 1.3 World Wide Web server.
 
  When will (if ever) HP and vendors come out of their tunnel and start
addressing global and public networking solutions?  I think the answer to
this question was expressed during the conference when George said: "We get
hundreds of requests for services but prioritized them by customer demands
for the services".
 
  You have to have requests for a market that you do not hold before you
take action?  How can you sell to educational and government institutions or
companies wanting to do business on the internet with software and vendors
that only work with private networks?  Who is doing password or data encryption?
 
 Well, you don't sell.  I guess you sell them HP9000's instead?  Right?  But
wait...
 
  I believe in the HP3000 more than any other machine platform for doing the
kind of reliable transaction updating characterized by the HP3000 line. I
know the evolution path of the HP3000 was small niches and such, but it
seems the hardware and MPE OS platform has overtaken the marketing and third
party vendor offerings, so much that portable platforms are being mapped
upon 3000's as a way to force HP3000 vendors to change into the open market
systems.
 
  The most used protocol on the internet never mentioned for HP client
server activity is the state-less type variety.
 
  Warning:  testimonial follows:
  --------
  Faced with the problem of delivering data to thousands of network
connected public and private machines (Pc's, Mac's, Unix), in 1993, we put
together an HP3000 gopher server that supports 4,000 campus access points.
 
  A worst case day for us was on Monday, Jan 16th, 1995.  We had flat rates
of 40 gopher network connections per minute over a 4 hour period while
averaging 9% usage of total CPU over a 12 hour period - delivering student
schedules and other reports accessing _Image_data_bases (real time stuff -
no copies).  Other times we did 20 connections per minute.  Nobody noticed
slowdowns on the 250 sessions logged on.  Why?
 
  First, the CPU wasn't max'd out.  That helps.  Second, this protocol is
"state-less".  What the heck is "state-less"?
 
  All the demos of client server ever witnessed on HP videos is stateful -
you log on and run stuff for a long time.  The servers for SQL access follow
the same principles.  While the user at the client is having fun playing
with their data, the logon remains connected taking up machine resources on
the host.
 
  The nature of state-less is that you promise more than you can deliver to
all user end points at any given time - like the phone system, like plumbing
in your building, like the highway you drive on..... Which scales the
delivery mechanism many many times over a logon method.
 
  You shoot for load _averages_, not absolute numbers.
 
  State-less protocols are characterized by short duration transactions,
that release resources to others while users are playing with data returned
on the client.  A typical transaction, for example, uses these four basic steps:
 
connect
request
respond
close.
 
  Ok, if you keep your transaction short, say 0.5-3 seconds, how many
customers can be served in this manner?  Well, that Monday for example, we
did a little under 24,000 socket connections in 12 hours while sometimes 250
sessions were logged on and nobody noticed except the Glance screen was full
of dead gopher processes in a 30 second sweep (not to worry, it is supposed
to work that way!)
 
  George kept saying "amazing" when he saw a SQL speedsheet transfer to a
PC.  BTW, we do speedsheet file transfers via HP gopher to Mac's, PC's, Unix
host machines too.
 
  I come to work and can't believe the number of accesses our students are
doing on our HP3000 Corporate data over our campus network (slow times are
like 1500 connections a day while peak times 24,000 connections).  Then we
look at system resource usage at the end of the month for HP gopher - a few
percent or less of total resources consumed.
 
  As long as you have average usage over long periods of time, things work
pretty well.
 
  Now that is truly amazing,  when thinking back 5 years ago about the state
of educational networking on the HP3000!
 
 
--------------------------------------------------------------------
Eric J. Schubert                 Administrative Information Services
Senior Data Base Analyst         University of Notre Dame, IN USA