Joe Geiser wrote:
> I configured a link (in NMMGR) for the lan, identical to that of our
> other 3000's.  If using Reflection/1 for Windows to our other 3000's,
> I don't lose the link.  The other 3000's are running 5.0.  However,
> the 928 is under 5.5 (remember, with an idential config in NMMGR),
> and it will timeout and kill the connection after varying amounts of
> inactivity.  There's no rhyme or reason behind the timings either ---
> it could be a minute, or it could go 4-5 minutes.

If all of your 3000 network activity is on a single physical segment
with strictly Ethernet connections, you probably haven't read this far.

If you're on a *busy* segment, maybe you're affected, though this seems
to be a 5.5 "thing".  Much of NS transport went native mode and
introduced some new quirks.

Bottom line, however, is a personal observation (lawyers back off!) that
the 3000's TCP stack is tuned to a local-area Ethernet with low traffic
and it simply does not stand up to heavy traffic (locally) or really
remote traffic (network latency, slow dial-up connections, etc).  I have
had less obvious problems with WRQ's RNS (on the client end) with dialup
connections.  They all scream locally at ethernet speed, which is the
"typical environment", but I'm not so sure that either has been tested
under "stressful" (real-world internet) conditions.

I've always "thought" this but never had any proof.  My recent "horror
story" of problems left me with a 486/66 Win95 PC/MS TCP in place of my
previous 386/33 Win3.1 RNS TCP. The latter suffered from timeouts (not
related to processor speed) and generally bad performance over the
network in general.  The new one behaves quite well except for telnet
and FTP to a 3000 - bad performance.

Real "internet" access doesn't seem to be mature, let alone optimized in
our arena.  I'm confident that it will improve, but after seeing telnet
performance over a PPP dialup line with plenty of CPU behind it, I'm
certain that the 3000 needs attention since other hosts exhibit no
problems.

Jeff Kell <[log in to unmask]>