IBM Releases GPLd WinModem Support For Linux

horst writes: "Subject says it all -- IBM has released first GPL winmodem driver. Link found at LWN" I'll be even more excited when they release the code that works with my T20 ... I've never even dialed my modem *sniff*, but if you've got an MWave (600, 600E, 770) then you should be golden. But props to IBM for making a cool move. Hopefully it's not an isolated one.

Comments

Are winmodems really that bad?

[snellac]

OK, so most Linux people don't like the fact that winmodems are closed devices that are not supported on Linux. The conventional wisdom complains that existing winmodems give poor performance and kill the CPU. However in a recent /. post no less a personage than John Carmack suggested that winmodems could be implemented in a way that is better than conventional modems for the needs of interactive games.

In the process of doing a web search I then turned up Stuart Cheshire's old home page. For those who don't know who he is, well before the web was popular he wrote a classic networked Mac game called bolo. (In fact when the web became popular the bolo players used to curse that the web was dragging the internet down too much...) Most links to it are dead, but the official home page is still up although there has not been a release since 1995. (This was apparently done as research into the needs of interactive networked programs. Gee, all of those hours that I spent as a test subject without knowing it...)

With Stuart's credentials established, it is well worth looking at his rants. In particular his latency rant, which was expanded out into a white paper.

Once you are through reading those you will see that for anything interactive, particularly games, what really matters is latency, not bandwidth. And modems are a major source of this latency. In addition he and John Carmack agree that software modems (AKA winmodems) can be (though they are not currently) programmed to operate in a mode that reduces latency, and the result would be better for interactive games than conventional modems.

So, are winmodems just a bad idea, or are they just poorly implemented? Conventional wisdom says that they are bad no matter what. But the people who should know best suggest otherwise.

-snellac

Re:Are winmodems really that bad?

[karmawarrior]

So, are winmodems just a bad idea, or are they just poorly implemented? Conventional wisdom says that they are bad no matter what. But the people who should know best suggest otherwise.

There are two types of Winmodem, modems that have most of the logic in the "PC space" of the modem, and those that have most of the logic in the "Modem" space of the modem. Usually, the logic is in the "PC space", that's why they're cheap - instead of having an expensive DSP do the work of turning raw PCM into V.32/34/90/22/etc/*, the Pentium does it. This is cheap. It also means that the performance of the PC itself suffers, but adds as small compensation the fact that the data doesn't have to flow from the modem to the PC post-decoding, ie it's ready for the PC to use right away, reducing latency.

In IBM's case, this model is not what's being used. The MWave is a DSP. So the "cheap" argument goes out of the window, except in that IBM recognised that a DSP could be used for multiple applications, and originally the MWave chipset was implemented by them exactly that way. My TP 760XD for instance uses the MWave chipset to provide both modem functionality and 16 bit soundcard support. Latency isn't likely to be a problem as the communications between the DSP and the "PC space" is much tighter than it is with a conventional modem, which usually goes via a real or imagined serial link controlled by a conventional UART chipset. So latency is going to be better than it would be with a real modem, but not as good as it would be with a conventional Winmodem.

Essentially you could say there are three types of modem: Conventional, open, serial modems, which will work with everything at a minor latency tradeoff, Winmodems, which will only work with the operating systems (or rather system, support for non DOS Windows based operating systems is rare, and that includes other Microsoft operating systems such as NT) supported by the manufacturer, and will slow down your computer's performance with a small advantage in the latency stakes, and Other Proprietry Modems, such as the IBM MWave set-up, where you still have the problem that the OS has to be supported by the manufacturer, but neither reduced performance or latency are real issues.

On the face of it, if someone could invent a generic device driver mechanism, or even just force, somehow, manufacturers to produce open source drivers, IBM's approach would probably be quite good. As it is, a year or so after IBM started this project we have a driver that only addresses the modem side of the MWave and only works with the later, less popular, Thinkpads. I'd have rather they worked on the soundcard functionality, a good PCMCIA Modem costs less than $30 on eBay these days. Grumble.
--
Keep attacking good things as "communist"

Double correction

[mindstrm]

Those 'dsp' files are firmware to be loaded into the modem card itself, and processed onboard. There is no reason we need source for these, and the same files would be used no matter what OS it is. The trick is how to get the contents of those file sinto the modem so the DSP on the modem can use them.

They aren't even technically part of the 'driver'.

You have the freedom to make their modem work on *any* hardware platform now; just not to steal their DSP code.

Re:Double correction

[Anonymous Coward]

The problem is that if you modify the dsp sources your modem ceases to be legally usable on the public telephone network. So what would happen is, since most people don't have access to telephone line emulators the drivers would have to be tested illegally. As far as I'm aware in every country it is illegal to attach unapproved equipment to the telephone network. So using the drivers would be illegal, and thus no distribution would risk the wrath of the FCC, and other governments. You wouldn't have any drivers at all if the dsp sources could be modified.

They never will

[Russ Nelson]

Every modem manufacturer I've talked to refuses to help with an Open Source driver because it's too easily modified. It's illegal to connect non-certified equipment to the public telephone network. These manufacturers don't want to participate in any way with something that's illegal. There's just no benefit to them.
-russ