Path: utzoo!censor!geac!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!sdd.hp.com!hplabs!pyramid!lstowell
From: lstowell@pyrnova.pyramid.com (Lon Stowell)
Newsgroups: comp.dcom.modems
Subject: Re: New Modems
Message-ID: <136557@pyramid.pyramid.com>
Date: 4 Dec 90 22:10:18 GMT
Sender: daemon@pyramid.pyramid.com
Reply-To: lstowell@pyrnova.pyramid.com (Lon Stowell)
Organization: Pyramid Technology Corp., Mountain View, CA
Lines: 53

>In article <1990Nov29.040945.23924@mthvax.cs.miami.edu> max@mthvax.cs.miami.edu (Max Southall) writes:
>>forced out too. Maybe we should ask Jerry Pournelle to do a *real* test!
A major problem with typical "modem tests" is that the tester
personell do not have knowledge of real world phone networks and
the types of impairments that actually exist.  An almost as
common failure is lack of suitable protocol analysis equipment
to determine whether failed connections are the fault of the
modem or the DTE (and driving software) in the test.

The EIA test suites address PART of the first issue...lack of
Telco knowledge, but IMHO very inadequately.   

The second part DTE vs DCE requires expertise, as the techniques
many modems use to STAY on an impaired connection will actually
trigger (improperly written) disconnections on the part of the
external machines used for the testing.  Many so-called faulures
are actually mismatches between the computer and the modems
techniques.

Use of a "reference" modem to guide results interpretations is
HIGHLY recommended....but gaining a scientifically reliable
reference point requires hundreds of man hours...as direct
head-head comparisons on known impaired lines with a
statistically significant sample are required....as 10
successive dial attempts between any given pair of end points
are fairly likely to traverse 10 differing physical routes...and
each attempt requires analysis to determine whether the modems
or the DTE's caused the disconnect--and even then whether it was
the modem's or DTE's (software typically) fault.

In qualification testing at a prior employer for international
V.32 modems, NONE of the Rockwell chipset units could compete on
satellite circuits (simulated and actual) with the proprietary
DSP implementations from the international vendors.  This is NOT
a negative comment on the Rockwell chips tho...because the
variation between the best of the Rockwell based units and the
proprietary DSP units was very slight...and on connections more
typical of using barbed wire or railroad tracks as circuit
carriers...

The more non-intuitive results were between differing vendors
using the SAME rev. level Rockwell chips... almost orders of
magnitude difference in thru-put or connectivity on impaired
ckts...   Largely those vendors with strong backgrounds or
engineers with real-world telco experience had far better
products than those who learned phone lines by using a simulator
or a textbook.  

ALL of the vendors used EXACTLY the same telco Simulator and
series of tests...it is just that the use of a Telco network
Simulator requires considerable expertise and
interpretation...skills sadly lacking in many modem (AND P.C
testing magazines..IMHO) vendors.