A modem for a PC may be either internal or external. The internal one is installed inside of your PC (you must remove screws, etc. to install it) and the external one just plugs into a serial port connector on a PC. Internal modems are less expensive, are less likely to to suffer data loss due to buffer overrun, usually use less electricity, and use up no space on your desk.
External modems are usually easier to install and usually require less configuration. They have lights which may give you a clue as to what is happening and aid in troubleshooting. The fact that the serial port and modem can be physically separated also aids in troubleshooting. External modems are easy to move to another computer.
Unfortunately most external modems have no switch to turn off the power supply when not in use and thus are likely to consume a little electricity even when turned off (unless you unplug the power supply from the wall). Each watt they draw costs you about $1/yr. Another possible disadvantage of an external is that you will be forced to use an existing serial port which may not support a speed of over 115,200 bps (although as of late 1998 and late 2000 most new internal modems don't either --but some do). If a new internal modem had say a 16650 UART it would put less load on the CPU.
Internal modems present a special problem for Linux, but will work just as well as external modems provided you avoid the high percentage of them that will work only for MS Windows, and also provided that you spend time (sometimes a lot of time) to configure them correctly. Some of the modems which will work only under MS Windows are, unfortunately, not labeled as such. If you buy a new one, make sure that you can return it for a refund if it will not work under Linux.
While most new modems are plug-and-play you have various ways to deal with the PnP configuring:
There are many Linux users that say that it's a lot simpler just to get an external modem and plug it in. But since new peripherals are PnP today, you will sometime need to deal with it, so why delay the inevitable? Still, the most expedient (and expensive) solution is an external modem (if you have a free serial port).
Many external modems are labeled "Plug and Play" (PnP) but they should all work fine as non-PnP modems. While the serial port itself may need to be configured (IRQ number and IO address) unless the default configuration is OK an external modem uses no such IRQ/IO configuration. You just plug the modem into the serial port. Since you usually plug the modem into a serial port (and connect it to power).
How can an external modem be called PnP since it can't be configured by PnP? Well, it has a special PnP identification built into it that can be read (thru the serial port) by a PnP operating system. Such an operating system would then know that you have a modem on a certain port and would also know the model number. Then you might not need to configure application programs by telling them what port the modem is on (such as /dev/ttyS2 or COM3). But since you don't have such a PnP operating system you will need to configure your application program manually by giving it the /dev id (such as /dev/ttyS2).
Connecting an external modem is simple compared to connecting most other devices to a serial port that require various types of "null modem" cables (which will not work for modems). Modems use straight through cable, with no pins crossed over. Most computer stores should have this. Make sure you get the correct gender and number of pins. Hook up your modem to one of your serial ports. If you are willing to accept the default IRQ and IO address of the port you connect it to, then you are ready to start your communication program and configure the modem itself.
An internal modem is installed in a PC by taking off the cover of the PC and inserting the modem card into a vacant slot on the motherboard. There are modems for the ISA slots and others for the PCI slots. Some new PC don't have any ISA slots. While external modems plug into the serial port (via a short cable) the internal modems have the serial port built into the modem. In other words, the modem card is both a serial port and a modem.
Setting the IO address and IRQ for a serial port was formerly done by jumpers on the card. These are little black rectangular "cubes" about 5x4x2 mm in size which push in over pins on the card. Plug-and-Play modems (actually the serial port part of the modems) don't use jumpers for setting these but instead are configured by sending configuration commands to them over the bus inside the computer. Such configuration commands can be sent by a PnP BIOS, by the isapnp program (for the ISA bus only), or by newer serial device drivers for certain modems. Under Linux you have a choice of how to configure the ones that don't get io-irq configured by the serial driver.
Software modems turn over some (or even almost all) of the work of the modem to the main processor (CPU) chip of your computer (such as a Pentium chip). This requires special software (a modem driver) to do the job. Until late 1999, such software was released only for MS Windows and wouldn't work with Linux. Even worse was that the maker of the modem kept the interface to the modem secret so that no one could write a Linux driver for it (even though a few volunteers were willing to write Linux drivers). With few exceptions, this is still true today (late 2000). Also, there is no standard interface so that different brands/models of software-modems need different drivers (unless the different brands/models happen to use the same chipset internally).
A third name for a software modem (used by MS) is "driver-based modem". The conventional hardware-based modem (that works with Linux) doesn't need a modem driver (but does use the Linux serial driver) After about mid-1998 most new internal modems were winmodems and would work only for MS Windows.
Finally in late 1999 two software-based modems appeared that could work under Linux and were sometimes called "linmodems". Lucent Technologies (LT) unofficially released a Linux binary-only code to support its PCI modems. PC-TEL (includes "Zoltrix") introduced a new software-based modem for Linux. There is a GPL'ed driver being developed for the Modem Silicon Operation MD563X HaM chipset (nee Ambient division of Cirrus Logic). Will other companies follow these leads and thus create "linmodems"? For a list of modems which work/don't_work under Linux see modem list. Links to "linmodem" drivers may also be found there. A project to get winmodems to work under Linux is at http://linmodems.org. They also have a mailing list.
There is some effort underway at reverse-engineering with at least one report of a winmodem that has been made to work under Linux (but not yet with full functionality). So by the time you read this there may be more linmodems.
For details of how to get some winmodems to work under Linux see the Linmodem-HOWTO (and/or Winmodems-and-Linux-HOWTO which is not as well written). If code is made available to operate a "winmodem" under Linux, then one may call it a "linmodem". Is it still a "winmodem"? Well, it's still a software-based modem. The term "Winmodem" is also a trademark for a certain model of "winmodem".
There are two basic types of software modems. In one type the software does almost all of the work. The other is where the software only does the "control" operations (which is everything except processing the digital waveshapes --to be explained later). Since the hardware doesn't do the control it's called a "controllerless" modem. The first type is a all-software modem (sometimes just called a software modem).
For both of these types there must be analog hardware in the modem to generate an electrical waveshape to send out the phone line. It's generated from a digital signal (which is sort of a "digital waveshape"). It's something like the digital electronics creates a lot of discrete points on graph paper and then the modem draws a smooth curve thru them. There must also be hardware to convert the incoming waveshape to digital. Then this digital waveshape must be converted to a data byte stream. The modem can't just send this data byte stream to the PC but must first do decompression, error correction, and convert from serial to the parallel bus of the computer.
The difference between the two types of software-based modems is where these digital waveshapes are processed (generated and interpreted). In the all-software modem this waveshape processing is done in the CPU using a Host Signal Processor (HSP). In the controllerless modem it's done in the modem but all other digital work is done by the CPU (data compression, AT commands, etc.) For example the Rockwell HCF (Host Controlled Family) does this. If the software that does these tasks could be ported to Linux and then there wouldn't be a major problem.
How do you determine if an internal modem will work under Linux? First see if the name, description of it, or even the name of the MS Windows driver for it indicates it's a software modem: HSP, HCF, HSF, controllerless, host-controlled, host-based, and soft-... modem. If it's one of these modem it will only work for the few cases (so far) where a Linux driver is available.
If you don't know the model of the modem and you also have Windows on your Linux PC, click on the "Modem" icon in the "Control Panel". Then check out the modem list (see Web Sites. If the above doesn't work (or isn't feasible), you can look at the package it came in (or a manual) find the section on the package that says something like "Minimum System Requirements" or just "System Requirements". It may be in fine print. Read it closely. If Windows or a Pentium CPU is listed as one of the requirements then it is not too likely to work under Linux.
Otherwise, it may work under Linux if it fails to state explicitly that you must have Windows. By saying it's "designed for Windows" it may only mean that it fully supports Microsoft's plug-and-play which is OK since Linux uses the same plug-and-play specs (but it's harder to configure under Linux). Being "designed for Windows" thus gives no clue as to whether or not it will work under Linux. You might check the Website of the manufacturer or inquire via email. I once saw a web-page that specifically stated that one model worked under Linux while implying that another model didn't.
Only if you know there is a Linux driver for it that works OK. Besides the problems of getting a driver, what are the pros and cons of software modems? Since the software modem uses the CPU to do much of its work, the software modem requires less on-board electronics and thus costs less. At the same time, the CPU is heavily loaded by the modem which may result in slower operation. This is especially true if other CPU-intensive tasks are running at the same time the modem is being used. Of course when you're not using the software modem there is no degradation in performance at all.
Is the cost savings worth it? In some cases yes, especially if you seldom use the modem and/or are not running any other CPU intensive tasks when the modem is in use. Thus there are cases where use of a software modem is economically justified. The savings in modem cost could be used for a better CPU which would speed things up a little. But the on-board electronics of a modem can do the job more efficiently than a general purpose CPU (except that it's not efficient when it's not in use). So if you use the modem a lot it's probably better to avoid software modems (and then you can use a less powerful CPU :-).
A PCI modem card is one which inserts into a PCI-bus slot on the motherboard of a PC. While most PCI winmodems will not work under Linux (no driver available) other PCI modems mostly work under Linux. The Linux serial driver is being modified to support certain PCI modem cards (but not winmodems). If the Linux serial driver supports it then the driver will set up the PnP configuration for you. See PCI Bus Support Underway If no special support is in the Linux serial driver but it may still work OK but you have to do some work to configure it.
Such modems use DSP's (Digital Signal Processors) which are programmed by driver which must be downloaded from the hard disk to the DSP's memory just before using the modem. Unfortunately, such downloading is normally done by Dos/Windows programs (which doesn't work for Linux). But there has been substantial success in getting some of these modems to work with Linux. For example, there is a Linux driver available to run a Lucent (DSP) modem.
Ordinary modems that work fine with Linux (without needing a driver for the modem) often have a DSP too (and may mention this on the packaging), but the program that runs the DSP is stored inside the modem. This is not a "DSP modem" in the sense of this section. An example of a DSP modem is IBM's Aptiva MWAVE.
If a DSP modem modem simulates a serial port, then it may be usable with Linux provided you're willing/able to boot from DOS. You must have Dos/Windows on the same PC. You first install the driver under DOS (using DOS and not Window drivers). Then start Dos/Windows and start the driver for the modem so as to program the DSP. Then without turning off the computer, go into Linux.
One may write a "batch" file (actually a script) to do this. Here is an example but you must modify it to suit your situation.
rem mwave is a batch file supplied by the modem maker
call c:\mww\dll\mwave start
rem loadlin.exe is a DOS program that will boot Linux from DOS (See
rem Config-HOWTO).
c:\linux\loadlin f:\vmlinuz root=/dev/hda3 ro
One may create an icon for the Window's desktop which points to such a batch file and set the icon properties to "Run in MSDOS Mode". Then by clicking on this icon one sets up the modem and goes to Linux. Another possible way to boot Linux from DOS is to press CTRL-ALT-DEL and tell it to reboot (assuming that you have set things up so that you can boot directly into Linux). The modem remains on the same com port (same IO address) that it used under DOS.
The Newcom ifx modem needs a small kernel patch to work correctly since its simulation of a serial port is non-standard. The patch and other info for using this modem with Linux is at http://maalox.pharmacy.ohio-state.edu/~ejolson/linux/newcom.html.
Some older Rockwell chips need Rockwell RPI (Rockwell Protocol Interface) drivers. They can still be used with Linux even though the driver software works only under MS Windows. This is because the MS Windows software which you don't have does only compression and error correction. If you are willing to operate the modem without compression and error correction then it's feasible to use it with Linux. To do this you will need to disable RPI by sending the modem (via the initialization string) a "RPI disable" command each time you power on your modem. On my modem this command was +H0. Not having data compression available makes it slower to get webpages but is just as fast when downloading files that are already compressed.