skip to the Modem Update
skip to the Modem Update on the Old 3Com Sportster Modems
a "Fix" for Hung 3-Com/USR Modems in a Hunt Group
OTHER PAGES
SCN Help Page
SCN FAQ (Frequently Asked Questions)
Terminal Settings and Telnet Programs
(Modem Init. Strings) - Max Speed - explains optimum modem speed
back toSCN's Dialup Connect Speeds

NOTE: This document is here only for historical documentation reasons.

Login Carriage Return

It is necessary to press the carriage return key once after your modem reports "CONNECT". Everything else follows the prompts.

SCN Modem Update

Brief Status

We believe, as of 17-11-98 we cured the gibberish problem on the Microcom modems (main hunt group). As of 20-11-98 we cured the gibberish problem on the USR modems. We have not cured the gibberish on the Internet. Users who want gibberish need to go to the Web.

The new modems are Compaq Microcom 4000 analog modems. The rack is also capable of accepting digital and channelized digital/analog modems.

We have, as of 28th Oct, 98 installed these as the first 18 modems on our 32 line hunt group. If you are calling at times other than 1800-2200, that's what you connect through. If you want to connect through a Sportster modem, dial 206-654-7044.

Detailed Status

We had the luxury of getting these right (in our system) before putting them on-line. Nevertheless, it took the better part of a day to get rid of all sorts of last minute bugs. Hey, it's not like sticking a modem in a PC and running Windows "Find my new hardware for me." These things go through Chase IOLAN terminal servers, which connect through another server to our Sun equipment. Then our Login program (Unix) has to recognize the incoming call, and communicate this using a telnet protocol which communicates back through the individual ports to the modems.

The gibberish is an example of things which can go wrong with a new connection. We set DTE speed to 57600, which is optimum for a 33.6 connection on this type of system. The modems for the most part would follow DTE, but on reset they would sometimes go to the default DTE speed of 115200, and we found at least two at 38400. At this point we had the choice of setting the hubservers to the higher speed or writing the 57600 speed directly to the modems. Since the 57600 speed was optimal, we chose to do that.

The Microcom modems have some nice features. Two useful features are an electronic "cold boot" capability and an ability to "busy out" the modem. Normally these commands don't take effect until a user disconnects, which is convenient. This sort of thing allows remote servicing of the modems and verification of smooth operation.

One feature we will investigate is an ability to adjust the modem's sensitivity to line noise. If we make it more sensitive, it reduces connect speed, but this actually increases data transfer speed. See Modem Settings - Maximum Speed for an explanation. Preliminary tests were favorable, but I want to make sure the settings really do what I think they're doing.

The last 14 lines are connected through 3-Com Sportsters. These get cold booted on a nightly basis so that if one hangs, it's self-correcting. We plan to provide "backup modems" so that when the Sportsters hang, a second modem will pick up the call on the second ring.

We are also still working out a wrinkle or two, such as a possibility that the modems are cutting off callers with 1200 baud modems. We actually have 20 rack mounted modems, but will be using the last two for maintenance and testing.


The Old Modems

USR/3Com Sportsters

Gibberish

The gibberish issue related to two changes we made to port baud rates. The USR modems set their port speeds in response to AT signals. Before that they look at NVRAM.. We resolved the problem by simply rewriting to NVRAM. We missed this last week because we believed (correctly) that the gibberish was coming from the Microcom modem connections. (Same problem; different setting.)

Ring/No Answer Fix

We found that during the "Ring/No Answer" hang, the modem's telephone extension port is live. By simply connecting a second modem to that telephone jack, the second modem is able to answer when the first one hangs. We have 20 of the affected USR modems and room for 20 extra port connections. Right now the first modems on the second hunt group are backed up in this manner, and we expect to back up all affected modems.

Since the Sportsters start with Line 19, a hung modem is not likely to cause major problems on the main hunt group. Those paying close attention to their dialup connections will note an occasional connection where the modem takes two to four rings to answer, followed by a 14.4 connection. That's our backup modem picking up the line.

The nightly shutdown event will then reset the modem by cold booting those modems.

Ring/No Answer

(not to be confused with a 'CONNECT/blank' in which SCN fails to provide a prompt.)

We believe we have tracked this down to a Sportster Negotiation Hang. We are presently testing the modems to determine if disabling the 3429 symbol rate "disables" the hung state. For more details, see below.

Meanwhile our nightly cold boot will continue until we have determined that we can keep these pieces of '3Com' from hanging.

Unanswered Calls

There are two reasons calls are unanswered:

A modem hangs.
We believe that 3-Com/U.S. Robotic's single processor design causes their Sportster modems to hang.
Transcient Hangs
As with the case of a modem hanging, a subsequent call will get through to a different modem. We also believe that a subsequent call to the same modem will get through. When we were running just the 3-Com modems, if one modem is hung and two are transcient, it would take as many as six attempts to get in.

Daily Shutdown Event

Except on the second hunt group, the nightly cold boot event is a non-event. That's because at 04:00, we will never have more than 5 users (normally 2) and the first 3Com modem is on line 19. The 18 rack mounted modems DON'T get cold booted.

Besides, at 04:00, we figure people won't notice or care if the computer stops responding anyway.

Using JJ's industrial timer, we turn off the modems' power for 15 minutes. The outside connections, including SPL, will be unaffected by this.

The purpose of the cold boot is to automatically reset "hung" 3-Com Sportster modems that do not respond to a soft boot (ATZ).

Keyboard Pauses

Some users had noticed pauses of up to one minute. The appearance is that the computer locks up and then suddenly returns to life. We believe that we corrected this. See Pausing Bug (below, this page).

U.S. Robotics Initialization String

The best one we found is at www.allegro.com/sportster.html

What we've done so far regarding the dialup problems.

There are various problems affecting the dialup lines, not necessarily 'modem' problems. These problems have been difficult to diagnose because:

  1. They were not cases of individual modems going 'bad', but transient problems.
  2. There are a great many subtle variables, not all of which are known, and it is necessary to look at combinations of variables.
  3. Some of the problems have a time or usage dependent factor, so several days are necessary to test a change.
  4. The (3-Com/USR) modems do not have any kind of debugging or monitor mode.
We had:


These are the results of several hundred hours of research, experimenting, monitoring, and otherwise trying to pin down the problems, and a good many other hours trying to devise scenarios of what is happening. Some of the problems are not clearly distinguished. e.g., 'ring/no-answer' can be caused in several ways, and fixing one cause would reduce the problem, but not eliminate it. And the perception of various problems is highly variable, being dependent on specific circumstances. (Amazingly, a few users don't know what we are talking about because they have never noticed these problems! We should all be so lucky.)

The most serious dial-up problem currently is the "ring/no-answer" problem. This involves modems going non-responsive, and had required a trip on-site to power cycle the recalcitrant device.


Specific Problems

THESE PROBLEMS WERE WITH THE 3-COM/USR SPORTSTER MODEMS AND SHOULD PRETTY MUCH BE ELIMINATED.

Ring With No Answer

This takes the appearance of 'nobody' picking up the phone. At our end, the modem "sees no carrier, hears no carrier and says, 'NO CARRIER'". The effect is a phone line ringing with no answer from the modem. We noticed that the modems progressively go to a state where they do not recognize dialtones. 'ATZ' resets don't correct this, but a cold boot does. The cold boot event occurs between 04:00 and 04:16 with a possible 10 minute error. Incidently, with some settings, the response is 'NO DIALTONE'.

This hung condition occurred at the termination of perhaps 1 out of 2000 calls.

Gibberish (Modembarf)

We have occasionally seen modems go to a hung state which produces gibberish. We found this most often on modems which didn't have extra cooling slots cut in their cases and on those modems which were set to eliminate keyboard pauses. We found that by disabling V.42 error correction (s15=128), we have stopped the gibberish.

Line Noise

Not much we can do about this one, folks. Some users with non-error correcting 2400 baud modems are getting this, but our lines test out as clean voice quality lines. This testing was done at USR's 56K compatibility bulletin board and by using one of our 56K-capable modems. (Before you ask, enabling 56K does not work.) Their compatibility bulletin board is history, since they ended up paying for a WATS line so people could run frequency tests on their phones.

V.42

Pausing

The 3-Com/USR Sportster Pausing Bug, affected five of our 33.6 modems (PROM rev. 2.31). This is believed to be caused by a particular implementation of V.42 file compression, and was identified as occurring on those five modems. There are two patches for this; disabling V.42 file compression (s15.7=1 or s15=128), and disabling escape string guard time (s12=0). USR replaced these modems.

Some of the shorter pauses are caused by such things at the internet taking a 'convenience stop" so to speak, and these problems are of course not addressed by modem fixes.

Stuttering Dialtones

We have suspicions that some problems are caused by the nature of our phone connections with U.S.West. Specifically we had discovered that U.S.West had slammed us with three-way calling. (Three-way calling does not generate a true stuttering dialtone, but does affect connections when the line is placed on-hook for a short time period.)

This "feature." has been removed, as we suspect that it could cause problems with our modem connections. These include unanswered calls and one-time unanswered calls caused by slow resets.

"Scrollin' Sadie" Gone but Not Forgotten

We believe we have the "Scrollin' Sadie" problem solved. This was seen as a scrolling series of login prompts, often in an ordered pattern. It made a great screen saver, but was needless to say frustrating to users, who often had to manually intervene to interrupt the sequence. In some cases, it would cause Telix and other terminal programs to lock-up.

The scrolling was caused by the modems spontaneously changing their settings, which I think occured at the termination of a session. We found that we can configure the modems to reset at the termination of a session, so if the modem changes its setting, it will only do so for a single call.

When we tried this before, we lost one hub (needed to be rebooted), but we think that this may be the result of the way that the 14.4's reset. The spontaneous settings only affect the 33.6's so our new configuration will only apply to the 33.6's.

We believe we also found the cause, which is generally related to either the '+++' escape code used by some software as a backup to dropping DTR when a user hangs up or redials, or possibly line noise inherent in some 2400 connections. While we weren't able to force scrolling in this way (we didn't try very hard), our modems no longer respond to an escape code. Therefore, even a first instance of "Scrollin' Sadie" should occur less often.

Whatever the cause, this only occurs with our US Robotics single processor modems, and not with the 14.4's which are dual processor modems.

Specific Problems found with U.S. Robotics Modems

The following problems with U.S. Robotics Sportster Modems have either been discovered by us, or have been reported. Not all of them affect us and not all of them apply to our specific modems:

That said, U.S. Robotics chose an excellent design for the case. The wedge shape turns out to be very handy as a doorstop.



SCN Home
SCN Help Page
SCN FAQ (Frequently Asked Questions)
Modem Initialization Strings Terminal Settings and Telnet Programs

- Stan Protigal - Comments about this site: email me

SCN Home Page
Seattle Community Network