subscribe

Path: sun.sirius.com!usenet
From: Don Hurter
Newsgroups: sirius.tech
Subject: 28.8k modem update, Friday evening
Date: 16 Sep 1995 02:56:27 GMT
Organization: Sirius Connections
Lines: 142

Here's the situation; we have gone through a lot of false starts with 28.8k service, particularly after an extensive testing session earlier this year. Our experience with testing inexpensive 'stand-alone' modems, such as Sportsters and Supras was less than satisfying. None of these class of V.34s could reliably receive calls, maintain long-term connections, or provide any level of consistent performance, let alone lend themselves to any sort of physical manageability w.r.t. power supplies or cabling. In the meantime, we were using stand-alone USR couriers for dedicated dial-up clients, and they were rock solid. With this data available, we decided to try out USR MP16s, which gave very good results in tests parallel to the stand-alone units. They are rack-mounted boxes which contain 16 Courier chip-sets all concentrated onto two circuit boards, and cost quite a bit of money. Nonetheless they initially allowed us to offer what we assumed would be reliable 28.8k service, so we purchased three units to start out with.

The USRs performed quite well for the first month of operation, but then one unit locked up the lower eight-port card (all status lights glowing in the 'critical failure' status.) We power cycled the unit, which unfortunately cut off everyone connected to the still-functional upper card, and the modem reset itself. A week later it happened again, and we called USR to report the problem. They had never heard of this condition before, but promised to get back to us when they figured out the problem. Then a day later a different box locked up, and we called USR again, only to get put on hold for a few hours before someone said, 'Yeah, we're looking into it.' Unfortunately, that was the truth - USR had no more data to go on than we did. Intermittent problems that you can't repeat are damn near impossible to make any headway on.

Nonetheless, USR discovered that the connectors on the circuit boards themselves would unseat themselves after a period (they were trying to save money on their manufacturing costs.) They soldered up a few replacement boards and shipped the new boxes out to us, but by that time we had suffered over a month of sometimes daily modem failures, which ultimately kept our customers from connecting up.

During all this we explored other avenues for 28.8k modems, including Microcom rack-mounts and Multitechs. The modem business right now is a booming place for all manufacturers, and as a result units from either of those vendors took about a month to arrive. Then once we got them we had to configure them and test them internally before putting them online for customers to dial into, and even then we would put them at the end of the hunt group so that they weren't under constant load if there were any problems.

The Microcom racks looked promising from an engineering point of view - solid chassis with dual power supplies and hot-swapable modem cards. Unfortunately, the first unit arrived with a dead power supply (not field replaceable), so we had to wait another two weeks for a new unit. The Multitech modems worked right out of the box, and after testing we added them to the tail of the hunt group in August. These received calls properly, and appeared to hold connections. During all this, the replacement USRs started to fail again. In one case the boards still locked up, and in another two ports stopped working, causing no end of dial-in problems. USR remained aloof and wouldn't return our endless messages to the support number.

When the replacement Microcom arrived, it also had a dead power supply, and we also learned that connecting multiple phone lines to the units through their compact connectors would have caused us to re-wire hundreds of phone cables if we decided to go with them. At that point we were pretty disappointed with the whole thing, and instead focused our energies on the Multitechs. We ordered another 48 units, since we knew that by the time they arrived (3-4 week lead time in this case) we would max out the existing 28.8k modems.

A few weeks later another replacement USR failed, this time for our dedicated line customers. Fortunately a previously ordered replacement unit had just arrived, but that was slated for the 28.8k dial-up pool. This caused the start of the 28.8 busy signals a few weeks ago, while we eagerly awaited the second round of Multitechs. By now, however, we had decided to shit-can the USRs, as they had caused too much trouble for us and our customers.

At this point we had more than 50K and countless hours of unpacking, configuring, and testing hours tied up in these various modems, and still didn't have anything we were truly comfortable with for 28.8k service other than the sporadic USRs and one rack of Multitechs. We shipped back all of the Microcoms, if only to make some more room to work with everything else. The second round of Multitechs finally arrived, and we were none too anxious to get them online to fend off more busy signals. On Tuesday morning I jumped the hunt sequence around and took the USRs down to make room on the racks, and installed the Multitechs. We configured them identically to the earlier batch, and tested each one to make sure it picked up properly and reported the correct port address. Then we performed a few transfer and authentication tests, which they passed without any problems, and finally put them in the path of the 28.8k hunt sequence.

What happened after that was reported in an earlier message, and all week long we've been trying to pinpoint the source of the problems that everyone has reported. Today I talked yet again with the Multitech support person, and we think we have the problems worked out. However, the only way we'll ever know for sure is to put them back online (today I had all the 28.8k calls go to the USRs first while we hopped over the modems we were working on.) Unfortunately, you, the customers, will become the testing guinea pigs, as there is no other load test we can perform which will cause long term behavioral problems to appear.

As a side note, our phone lines have been quite stable over the past two months, with no dropped lines or other PacBell weirdness. However, just this week one of the phone lines plugged into the USRs developed a bad noise problem, and since i didn't have a full day to troubleshoot it with PacBell I just forwarded it to the next modem in the hunt group. Now if you were to guess when PacBell might take the opportunity to drop a line-forwarding at their Central Office, and you knew that they have left our service alone for over two months, and that we routinely forward calls every few days for maintenance reasons without any problems, what day out of the entire calender do you think the forward would fail? I'll leave that proof to the reader, but I'll tell you that that was the cause of busy signals last night when we actually had 30-odd idle modems patiently waiting for someone to talk to.

The weekend has come, and we have to wait until Monday for any additional support from Multitech if things still don't work out. Theoretically there should be enough modems to answer the calls, but for now I have no way of knowing how they will behave until they get used. We'll monitor their performance for the next few days, but can't perform any additional tuning or reconfiguration without running the risk of possibly making matters worse.

People call us a lot and ask for 28.8k modem recommendations. I've been following the comp.dcom.modem newsgroups, listen to stories from customers about other online services, have had accounts of my own on other services, and have had a mind-reeling experience with 28.8s here at Sirius. I truly wish V.34/28.8k modems would settle down so we didn't have to waste so much effort on them, but the reality is that modems in general are hacks dating back to ancient communication technology, and 28.8k modems are hacks-squared which push phone technology to limits that would make Alexander Bell scream in hallucinogenic terror. The best we can do is flog them along until ISDN becomes a commodity market, and even then a new set of problems will take over that I don't have the stomach to go into right now. Wide area communications is one of those technologies that suffers from teenage-like growing awkwardness, and the InfoSuperF-ingHighway is really just a scam dreamed up by copper foundries who were suffering from sales lulls after they stopped producing all those brass artillery shell casings in the Cold War days.

I'm joking, of course (but you can't be too sure :-), although I must confess that I've lost more sleep this past week dealing with those damn modems than I care to admit, and look forward along with everyone else to when we get everything sorted out.

-- Don