DSL ISPs Cable Companies Dialup Internet Service Providers
Special Report:
Boardwatch March '98 Cover
Story
Background
Boardwatch magazine stirred up controversy with a special cover
story in their March, 1998 issue. The story reports Boardwatch's
tests of x2 and K56flex modems and ISPs, and shows x2 massively
out-performing K56flex.
Opinions on the accuracy of the tests vary. Some find that the
results match their expectations, while others don't see any
connection between the tests and everyday experience.
Other real-world performance tests
One reason the Boardwatch results are so controversial is that
they don't agree with other published results. Boardwatch found
that the U.S. Robotics x2 Sportster averaged initial connect
speeds of about 45.2K. In contrast, the Hayes Accura's average
initial connect speed was an underwhelming 30.8K, roughly
two-thirds as fast as the Sportster.
Those numbers are startling because they're so far out of line
with other published, real-world tests.
In the 56K.COM
Reader Speed Survey, x2 modems had a slight advantage over
K56flex modems, but generally by only a few K in initial connect
speed, and a fraction of a kilobyte per second in download
performance.
In the last year, a number of leading magazines have tested
modems as part of comparative reviews. Their results show K56flex
and x2 modems neck and neck in terms of initial connect speed and
download performance.
Responses from the companies
involved
I am in the process of contacting a number of companies,
including 3Com, Boardwatch, Hayes, and Lucent. Their responses
will be added to the report.
Rockwell responded by forwarding a letter sent to the
Boardwatch editors:
- Rockwell's
response to the Boardwatch article - notes that Boardwatch
performed all of its tests from one location. Line impairments
at that location may have favored x2.
- Lucent's response - Lucent
responded by noting that Lucent-based K56flex modems took the
top three slots in C|Net/Computer.com's test of 56K
modems.
How do x2 modems report initial connect
speed?
The Boardwatch results relied heavily on initial connect speed.
Though most modem experts agree that initial connect speed is an
unreliable indicator of performance, the fact that almost all
modem software reports initial connect speed makes it a convenient
indicator, and one that most users rely on.
A source who wished to remain anonymous sent the following
letter, which reveals some interesting facts about the way x2
modems report connect speed:
The Boardwatch article is almost completely
worthless. It is largely based on the data rate reported in the
connect message. Unfortunately that isn't a valid number for
one of the two legacy 56K modulations. Prior to entering data
mode X2 modems estimate the data rate they will get (this is
probably done during phase 4 of the connect, that's where it
occurs in V.90 and K56flex). Most modems estimate the data rate
they should use and then switch to that rate. X2 estimates the
rate and then connects at 32Kbps.
The CONNECT XXXXX response prints the estimate not the
32Kbps value. After about 10 seconds the modem rate
renegotiates upward to the data rate it has determined can be
maintained from analyzing the line in data mode, this may be
the same as the rate estimated or higher or lower. So the value
in the connect response may have nothing to do with the actual
data rate achieved on the line.
From a technical view I prefer the idea of initially
connecting at 32Kbps. It prevents problems during the ADP/ODP
sequence when LAPM is starting. It allows the modem to rate
renegotiate upwards after actually measuring the line during
data phase instead of trying to estimate the exact data rate
initially. If every company did this in V.34 and V.90 I believe
consumers would be happier since their connections would be
more reliable.
Unfortunately consumers and modem companies like to brag
about high data rates. The only place they can see them is in
the connect response. Printing CONNECT 32000 isn't acceptable
even if it is true, so in order to keep the customers (and
marketing people) happy the modem estimates what data rate
might actually be possible and prints that. I suspect (but
don't want to take the time and effort to prove) the estimate
printed by X2 is usually higher than the actual data rate the
line will sustain.
The information about how X2 connects is semi-public
knowledge among the TR30.1 and ITU Q23/16 members, since some
of the 3Com developers have mentioned it in private
conversations. I first learned about this in December while
talking to a third company that was looking at what really
happened during an X2 connection, things didn't look right to
them. (I can't be more specific or someone might guess who the
third company is.)
Hall Stevenson wrote:
My ISP, erinet.com, has been supporting USR's x2
since it was released. Along with that, they created a web page
that can report their users "connection status." The address
is: http://home.erinet.com/cgibin/connection.cgi
(it may not work for non-users, in case you try it).
(It doesn't. - Les Jones) Here is a
copy-and-paste from my current session:
(Note that TX and RX are reversed, because the stats are
being reported from the ISP's perspective. - LJ)
Connection Properties Specific Connection Information
Your IP Address xxx.x.xxx.xxx
Initial TX Rate 32000
Initial RX Rate 24K
Current TX Rate 37333
Current RX Rate 24K
Modulation x2server
Transmit Freq 0 (hz)
Recieve Freq 1920 (hz)
RX Gain 1 (db)
Transmit Symbol Rate 2146
Recieve Symbol Rate 3200
Retrains Request 0
Retrains Granted 0
Bit Line Errors 172
Blocks Resent 25
Block Size 0
X2 Status x2Operational
Speed Downshifts 0
Speed Upshifts 0
Characters Lost 0
Characters Sent 2057105
Characters Received 240750
Error Control ccittV42SREJ
Data Compression ccittV42bis
As you can see, it reports an initial connection of 32k
and then steps up to 37k (hmmm, the Speed Upshifts reports (0)
upshifts???). I'm running Linux at the moment and the dialer
program I'm using, ezPPP, reports the following in its terminal
window:
CONNECT 45333/ARQ/x2/LAPM/V42BIS
Windows 95 Dial-Up Networking is basically the same...it
reports a higher speed than my ISP does.
In response, our anonymous source wrote the following, and
offered some suggestions for modem testing:
I believe the I11 and I6 reports on 3Com modems
report the actual data rates and don't include the estimate
printed as the connect rate. If true this makes it fairly easy
to check the actual rate during or after a call.
The only real way to test a modem is to measure
throughput vs. network coverage. This is what all the magazines
did for V.34, though most of them over simplified the tests. I
can understand their desire to simplify things, TSB-37 defines
168 test line combinations for V.34 modems for 100% network
coverage. Testing to 99% coverage (100 lines) with all the
different data transfer options (compressed, uncompressed,
single and bi-directional) takes about 22 hours for a single
V.34 modem. Unfortunately the network model for PCM modems
hasn't been published yet (TR30.3 is still working on it), so
there isn't a standard the magazines can test against. They are
going to like it less than TSB-37 since there are even more
line combinations defined, currently there are around 330.
A number of readers on the comp.dcom.modems
newsgroup have noticed that the initial connect speed of their
modems haven't matched the statistics provided by ATII6 and ATI11.
It appears that the ATI6 and ATI11 commands should be trusted
instead of the initial connect speed.
Found a mistake or a bad link? Contact us.
Copyright © 2008 Marchex, Inc. All Rights Reserved.
|
