(Forwarded back to HP3000-L...)

Evan Vaala wrote:
>
> Jeff, Do you mind giving me some feedback on this issue?

> [log in to unmask] wrote:
> > Jeff Kell of Univ. of Tennessee has also been receiving these telnet
> > update patches. He's the one you should speak to about telnet
> > efficiencies on a large-user-number machine. It's my impression
> > that he's been pleasantly surprised.

I don't know if I have the "latest-and-greatest" patches, but yes, we
do have some nonstandard brew of telnet.  Here's a bit of background
to illustrate where I'm coming from.

We have our highest session counts during the Fall and Spring start
of semesters when there is fee payment, late registration, parking
decal sales, pay-your-parking-ticket lines, etc.  Until this spring,
we were still running on a 3000/960.  We would have upwards of 160
users.  Until recently, most of these users were coming in serially
through a DTC (buildings not yet networked) but the past few years
they have been switching to telnet as the network grew.  Initially
we had users coming through a Telnet Access Card (TAC) in a DTC, and
we noticed little change from serial.  As the network continued to
grow, the telnet users increased until we reached the 40-user limit
of the TAC.

At this point, we enabled host-based telnet.  Through DNS trickery, we
would wait until the TAC was almost full, then point the address to the
host for the overflow.  You could tell a difference once you had enough
host-based users logged on, and it pushed us into the "knee" of the
performance curve, exaggerating the overall effects.

Shortly after this past fall semester started, we received our new
3000-969/120 and consolidated the Library (950) and our 960 onto it.
At this past spring start of semester, we had over 180 sessions, at
least 95% of which were host-based telnet.  And it ran very well
indeed!

In my opinion, unless you have a very old/slow (935/950/955/960) system
and/or very memory constrained, you won't notice the difference between
host-based telnet, DTC telnet, or serial connections in terms of system
performance.  I would even venture to say that if you go with NS/VT you
would burden your system further since NS/VT requires the requisite
VTSERVER process for each session, and this process will be riding
alongside your application for the duration of the session.  Host-based
telnet has no such excess baggage.  Users with really high session
counts might be constrained by the maximum number of processes would
benefit greatly from using telnet rather than NS/VT.

Jeff Kell <[log in to unmask]>