Issue 3-15, April 15, 1998

Be Engineering Insights: Booting Variations For PC And LILO

By Rico Tudor

While all computer users know how to boot their machine, very few know what is happening under the hood. Depending on your operating system, you may see your machine reboot a dozen times a day, so often that some companies regard this period of time as valuable commercial property. For such systems, it may help to observe the congressional hearings before altering the appearance or function of your boot sequence.

BeOS users, however, should find this article sufficient. It explains booting in general, how the BeOS boot sequence works on the PC, how it can be customized for your machine, and demonstrates some common configurations.

As a starting point, I have chosen March 3, 1982. This is the date on the following file:

HIGHMEM  = 010000

    clr  r0
1:  mov  (r0)+, HIGHMEM-2(r0)
    cmp  r0, $end
    blo  1b

    add  $HIGHMEM, pc

    mov  $0172526, r0
    clr  (r0)
    mov  $-01000, -(r0)
    mov  $060003, -(r0)
1:  tstb (r0)
    bpl  1b
    clr  pc
end:

I wrote this bootstrap for the PDP-11. After entering the 18 machine words (in octal) using the front panel, 512 bytes would be loaded from 9-track tape, followed by Coherent, a UNIX-like operating system.

16 years later, nothing has really changed. Bootstrap code copies itself to a safe location, jumps there, and loads a larger bootstrap from some device. This is repeated until the operating system itself is loaded. In the modern world of code bloat, bootstraps are still written with an old skill: wringing the most work from the smallest program.

On the PC, the ROM-based BIOS acts as the first bootstrap, saves us the trouble of entering it by hand. After that, various boot stages are available. This article is most interested in LILO, written by Werner Almesberger. This package, and the BIOS itself, are essential to booting BeOS.

The following cases have been selected both to illustrate the workings of the boot process, and to provide practical setups for immediate use. I have tested each case personally.

Warning
Warning

THESE EXPERIMENTS MAY RENDER YOUR DISK UNBOOTABLE.

Before a change to your boot procedure, be sure it matches your disk partitioning and OS placement. Having a boot floppy for each OS is good protection, backups are better, and a scratch machine is best.

Case 1: BIOS -> LILO -> zbeos -> kernel

This is a schematic representation of the simplest boot sequence, and the one used by Release 3 BeOS. In detail, here are the stages:

  1. The BIOS. The code is stored in nonvolatile RAM, and is able to operate the basic peripherals of a PC. It loads the master boot record (MBR): this is sector 0 of the first hard drive (ignoring floppies for now). If supported by your BIOS, this can be a SCSI drive.

  2. The first stage LILO bootstrap. This is the content of the MBR. It employs the disk-reading services of the BIOS.

  3. The second stage LILO bootstrap. This is located within a Be file system. This stage is large enough to offer many features, like a boot menu, and can load large files. It, too, uses the BIOS to read disks.

  4. /system/zbeos. This stage prints the "Starting the BeOS boot sequence" message, does general initialization for the BeOS kernel. It can operate many more devices than the BIOS, display a fancy boot menu, and knows about the various Be-supported file systems. Life is pretty cozy at this point. It finishes by loading the BeOS kernel.

Stages (1) and (4) know where to find the next stage on the disk. How do the LILO stages, (2) and (3), know where to go? The answer is a utility in the LILO package called, confusingly, "lilo". To paraphrase Monty Python, it's spelled /bin/lilo, but it's pronounced "map installer." The map installer analyses each file needed during the boot process, which can be located on any of your Be file systems. The analysis yields the BIOS drive number and absolute sector for every sector of the file, and stores them as a big list in a "map file." The map file itself, as BIOS drive number and starting sector, is stashed in the first stage LILO bootstrap.

The major side-effect of this approach is the need to run the map installer whenever a file used in the boot process is moved. This includes moving a partition and changes to drive parameters in the BIOS configuration. The good news is the LILO bootstrap code needs no knowledge of the file system. Any type of file system supported by BeOS can host the program and map files.

The map installer is called from the shell command line, and operates according to instructions from a configuration file. Here is /etc/lilo/lilo.conf for Case 1:

boot = /dev/disk/ide/0/master/raw
  install = /etc/lilo/boot.b
  linear
  ignore-table

  image = /system/zbeos

Here is the command to install the map:

lilo -C /etc/lilo/lilo.conf

The name /dev/disk/ide/0/master/raw is BeOS parlance for the master IDE drive on the first interface. The "raw" device will ignore any partitions and target the MBR. The "linear" and "ignore-table" options are required at this time: the occasional "Invalid partition table" message can be ignored.

When run for the first time, "lilo" squirrels away a copy of the original MBR in /var/lilo/boot.0300. This can be restored using the following shell command:

cp /var/lilo/boot.0300 /dev/disk/ide/0/master/raw

Here is the list of files used by LILO:

/etc/lilo/lilo.conf
/etc/lilo/boot.b *
/etc/lilo/chain.b *
/var/lilo/map *
/var/lilo/boot.0300
/system/zbeos (and equivalent) *

If files marked by (*) are moved, remember to run the map installer again. If you don't, you'll hear about it.

So much for Case 1.

Case 2: BIOS -> LILO -> win95

In Case 1, LILO was used as a program loader, proceeding without delay from BIOS to BeOS. However, LILO can also serve as a boot manager, permitting you to select which OS to boot. As usual, you must prepare the configuration file, and run the map installer.

Here is lilo.conf for Case 2:

boot = /dev/disk/ide/0/master/raw
install = /etc/lilo/boot.b
delay = 50
linear
ignore-table

image = /system/zbeos
  label = beos

other = /dev/disk/ide/0/master/0_0
  loader = /etc/lilo/chain.b
  label = w95

I assume Windows 95 is on the first partition (0_0). The "label" option allows you to assign a word to each OS (try to control yourselves). The "delay" option gives you time to hit the Shift key, which starts the menu. Otherwise, LILO will make a wise decision on your behalf.

If 0_0 is not found during map install, invoke DriveSetup. Then try to mount the DOS partition. This should cause 0_0 to appear.

The "image" designation causes the entire program file to be loaded, and is compatible with zbeos, Linux kernels like vmlinuz, and certain stand alone programs (see Case 5). The "other" designation causes just one sector to be loaded, namely the partition boot sector. This is sector 0 of the specified partition, and it customarily contains an OS-specific bootstrap. Think of it as a per-partition MBR. If you're still awake, I'm sure you see distinct possibilities here.

Case 3: BIOS -> LILO -> LILO' -> linux

LILO comes from the Linux world, so we would expect the following to work:

boot = /dev/disk/ide/0/master/raw
install = /etc/lilo/boot.b
delay = 50
linear
ignore-table

image = /system/zbeos
  label = beos
other = /dev/disk/ide/0/master/0_0
  loader = /etc/lilo/chain.b
  label = linux

Or maybe not. Did your Linux administrator arrange to have a partition boot block for your Linux system? If not, this lilo.conf has made Linux unbootable. Tsk.

Case 4: BIOS -> LILO -> linux image

If Case 3 leaves you dazed and confused, it probably means LILO on the Linux side is running the MBR. However, we are running the MBR. There can be only one.

Rather than lop off each others heads, let BeOS run the MBR, copy the Linux kernel to our file system and add it to our list of bootable images:

boot = /dev/disk/ide/0/master/raw
install = /etc/lilo/boot.b
delay = 50
linear
ignore-table

image = /system/zbeos
  label = beos
image = /system/vmlinuz
  label = linux
  root = /dev/disk/ide/0/master/0_0
  read-only

I will not explain Linux options, except "root". /dev/disk/ide/0/master/0_0 is the equivalent of /dev/hda1 on Linux: the mapping should be obvious. Otherwise, use the Linux major/minor notation, e.g. 0x0301. You can even specify SCSI drives, so /dev/sda1 is expressible as 0x0801.

Case 5: BIOS -> LILO -> memtest86

Memtest86 is a cute stand alone program by Chris Brady which performs extended memory testing. It is available from the usual places.

boot = /dev/disk/ide/0/master/raw
install = /etc/lilo/boot.b
delay = 50
linear
ignore-table

image = /system/zbeos
  label = beos
image = /system/memtest86

When "label" is omitted, LILO creates something sensible.

Case 6: BIOS -> Partition Magic -> LILO -> zbeos -> kernel

We must now face the problem illuminated by Case 4: sometimes we cannot control the MBR. For those of you with commercial boot managers, like Partition Magic or System Commander, the MBR is in their control.

LILO's role, once again, is lowly program loader. The first stage LILO bootstrap, evicted from the MBR, must reside in a partition boot sector. The partition is determined automatically from the /boot file system.

install = /etc/lilo/boot.b
linear
ignore-table

image = /system/zbeos

You may override the assumed partition using the "boot" option, but make sure it belongs to BeOS. Otherwise, you create a spectacle.

Case 7: BIOS -> LILO -> LILO -> zbeos -> kernel

If you are too stingy for a commercial boot manager, and want to take the ideologically pure path, you are ready for a graduate course in LILO.

The plan is to use LILO as a boot manager, meaning it runs the MBR. A configuration file must be prepared and the map installer run.

We also use LILO as a program loader at the partition level. This calls for a different configuration file and map installation.

By separating these functions, we do not lose face like Linux in Case 3. The revolving door of upper management in the MBR suite has no effect on BeOS booting.

Here is "lilo.conf.partition":

boot = /dev/disk/ide/1/master/0_0
map = /var/lilo/map.partition
install = /etc/lilo/boot.b
linear
ignore-table

disk = /dev/disk/ide/1/master/raw
  bios = 0x81
image = /system/zbeos
  label = beos

Here is "lilo.conf.mbr":

boot = /dev/disk/ide/0/master/raw
map = /var/lilo/map.mbr
install = /etc/lilo/boot.b
delay = 50
linear
ignore-table

other = /dev/disk/ide/1/master/0_0
  loader = /etc/lilo/chain.b
  label = lilo2

Here are the commands to install the maps. Note the order:

lilo -C /etc/lilo/lilo.conf.partition
lilo -C /etc/lilo/lilo.conf.mbr

The scenario has been enlivened by placing the Be file system on the second IDE interface. A bug in BeOS LILO will surface if your machine has this configuration, and a CD-ROM drive as slave on the first interface (/dev/disk/ide/0/slave/...). The bug relates to the BIOS drive numbering scheme. Fortunately, the "disk" and "bios" options allow you to specify the BIOS drive number explicitly: 0x80 is the first hard drive, 0x81 the second, and so on.

On a modern BIOS, SCSI hard drives are part of the numbering system. If your SCSI drive is the designated boot device, then it will be 0x80, and the other drives follow. CDROM drives are not numbered.

Needless to say, problems with BIOS drive numbering will be fixed.

Case 8: BIOS -> zbeos -> kernel

This last example makes no use of LILO, but comes in handy if earlier LILO experiments go horribly wrong.

cp /system/zbeos /dev/disk/floppy/raw

This shell command creates a bootable floppy...you can climb down from the window now.

Miscellany

For some, a penny saved is a penny earned: partitioning is unnecessary and wasteful. You can create a Be file system across the bare disk with a shell command like this:

mkbfs /dev/disk/ide/0/master/raw

The MBR exists, the partition boot sector doesn't. Implications are left as an exercise for the reader. Reader? Hello?

URL References:

ftp://sunsite.unc.edu/pub/Linux/system/boot/lilo/ ftp://sunsite.unc.edu/pub/Linux/system/hardware/memtest86-1.4.tar.gz


Be Engineering Insights: Bugs and You

By Steven Black

Greetings, I'm a new face in Newsletter articles. I'm Steven with a "v." (Not to be confused with Steve Sakoman or Stephen Beaulieu.)

I'm part of the long dark arm of Quality Assurance here at Be. We find bugs. We reproduce bugs. We investigate some of the bugs you submit. We don't fix them or assign priorities.

Given my rich experience with all sort of bugs, that's what I'll be talking about in today's article.

Welcome to Release 3 and Writing Better Bug Reports

While we do our best to hunt down bugs and clear them up, there are bound to be a few that escaped in our latest release. You can help out in the bug wars in a variety of ways. The first and foremost is by writing a good bug report.

Whether you're a registered developer who submits bugs via the special bug submission page on our web site, or you're a regular user who wants to see an annoying bug you found get squashed, your bug is important. To give your bug the best possible chance of being fixed, here are some guidelines to keep in mind when you submit a bug report.

  1. Make sure your bug is reproducible. Try it again on your hardware. Try it on someone else's hardware if you can. If you think it's related to a specific piece of hardware try to swap it out to verify, assuming that it was a hardware fault.

    Make sure the bug is reproducible when you've rebooted cleanly. (This would clarify that the death of the app_server when you used Telnet was actually caused by a third-party Foobar program that failed inelegantly in another Workspace, and it started gobbling up resources, which started a slow spiral of death, which happened to culminate when you ran Telnet that day, rather than just being a Telnet bug.)

  2. If the bug is reproducible, remember to include your hardware and software configuration information, including what version of BeOS you are running, and what other applications need to be running to reproduce it. Keep it simple, try to make the bug reproduce in as few steps as possible.

  3. Write the bug report in painfully clear and concise steps. Elaborate on everything and anything you think might help us reproduce your problem. Doug Wright spoke about this Newsletter Issue 101, Developer Workshop: Bring Your Own Bugs:

    www.be.com/aboutbe/benewsletter/Issue101.html#Workshop

  4. If you're getting a debug terminal, or you have serial debugging output enabled, you'll want to install your xMAP files. They're on the CDs in optional/xmaps. They'll also be on your hard disk if you installed the optional items, though they will still need to be installed.

    The kernel_*.xMAP file goes in /system, and the other xMAP files go in /system/lib. You'll need to reboot to make use of them. They're also useful in your own debugging, so they're good to install anyway. They're included on Release 3 for both PPC and Intel, and were on the PR2 CD. They're different for every release, so keep that in mind. They need to match the libraries, applications, and kernels you have available. If the xMAP file isn't current, delete it—it's less confusing than having an old one.

    How do you tell if an xMAP file is not current? Because of the way things are built and the fact that we do not date-stamp the world, xMAP files have slightly different times than the output file (shared libraries, applications, the kernel, etc.). Typically, the date-stamp of the xMAP files will be after the output file, though this might not always be the case. If you take them from the BeOS CD you installed from, they will be valid.

  5. If you're getting a debug terminal we could also happily use the output of the "sc" command. These are useful to us, regardless of whether xMAPs were installed. (It's just easier for us if the xMAPs were installed, 'tis all.)

  6. If you're getting dropped in to the kernel debugger (which I'll talk about shortly) we'll want the output of more commands. First, we'll want "sc", then "regs", and finish off with a nice "ps". This could get long, and most of the time we need only one if you've proven to yourself that it's a reproducible bug.

  7. To be sure you haven't overlooked anything, after you've typed in the bug, but before you've submitted it, reread it and add more information so that it's Reproducible, Clear, and Concise, and Has All the Proper Information.

Welcome to Kernel Debugging Land!

On rare occasions something leaks out about serial debugging. Serial debugging output is a low-level diagnostic/debugging tool used by the kernel engineers. It has all sorts of commands, all listed with the total available documentation on them, "help".

Requirements for serial debugging:

  • A BeOS machine to debug.

  • A machine capable of receiving serial communication at the appropriate settings.

  • Software that can communicate to serial ports at the appropriate settings, typically anything that could be used to connect to dial up BBSes and shell accounts (not necessary for dumb terminals). Connect works great if your serial output is going to another BeOS machine.

  • One of the following: a Mac serial cable, a null-modem cable, or serial cable with a null-modem adapter.

  • A strange bug or stranger interest which requires you to see the serial debug output.

Serial Port Settings:

On BeBoxes serial debugging goes out the serial4 port. On PPC machines serial debugging goes out the modem port. On Intel serial debugging goes out serial1.

  • 19200 baud

  • 8 bits

  • No parity

  • 1 Stop bit

  • No flow control

How to turn on serial output at boot time:

  • Make sure the machine that's going to receive serial debug output is turned on and has the appropriate software running.

  • Make sure your cables are connected.

  • Start up the BeOS machine you need to test.

  • Depending on the receiving hardware, do one of the following:

    • On BeBoxes press the F1 key and hold it, just after the BeOS logo starts to swirl.

    • On PPCs press the "delete" key and hold it, just after the BeOS logo starts to swirl. (Please note: This is the "delete" key that's labeled "backspace" on PC-style keyboards.)

    • On Intel press the F1 key and hold it after you first see the boot-loader message.

  • If you did it right, you should start seeing debug output, before you get the BeOS boot menu.

  • If you got the boot menu and no output, you either pressed the key too late, or (more likely) you don't have your cable or comm settings quite right.

  • If you didn't get the boot menu at all, you probably pressed the key too early. (You might also have a rare keyboard which doesn't send the key back as expected under this circumstance.) You could try breaking into the kernel debugger, but I recommend rebooting and trying again, as it's much easier to check your settings when you have serial debugging on and are in the boot menu.

  • To check to see that you are connected properly, Select "Rescan" from the menu. If you are currently connected properly, you should start receiving a stream of messages.

  • Select your boot volume and continue as normal.

Dropping into the kernel debugger

Provided that you have the appropriate cables already installed, (preferably verified that they are functioning properly), you can drop into the kernel debugger at any time.

Why drop in to the kernel debugger? A number of reasons, actually, including the following:

  • You want serial debug output, but you forgot to start it at boot time.

  • Your machine is hung, has died, or otherwise has ceased functioning, and you want to see what programs were running so you know who to blame.

  • You suspect your app_server has died, and have no way to Telnet in and verify. (This might be the case if your display wasn't updating, or had gone whacko, though it was still doing hard drive accesses.)

  • The serial debug output says something bad happened, and you want to gather some more information to send in your bug report.

How to drop in to the kernel debugger

On BeBoxes press the right-most button on the front of the case. On Power Macs press Command+Power On. On Intel machines press Alt+SysReq (also known as Alt+PrintScreen, this may change in the future).

What can you do in kernel debugging land?

  • ps - shows a process listing

  • run - just shows the currently running processes

  • regs - shows what the registers contain

  • sc - does a stack crawl on the current process.

  • c - continue executing/leave kernel debugging land

  • help - the documentation for kernel debugging land

Please note: Most, if not all, of the shell commands are designed for use by kernel engineers. The commands above are all you'll ever need to use, unless explicitly told otherwise by someone at Be.

Some of the available commands can make your machine lock up, lose data, and otherwise cause problems, so experiment at your own risk.

What we at Be want to see from your serial debug output

Typically, we only want the interesting bits of output. If something's repeated 10,000 times we might want to see one line of what it says, but we don't want to see all 10,000 repeated lines of it.

We are interested in all conditions that cause the machine to lock up, entirely or partially.

For degrading conditions or high debug-output conditions, we are interested in the point at which the output turns from good-normal debug output to the error output.

If the output starts streaming quickly and you know the machine is dying, break into the debugger as soon as possible. This allows you to show us the boundary between the good/bad debug output. (Please note, however, that it's normal to get a lot of output while booting, or while rescanning the device buses.)

Most of the time, giving us the last five lines of the debug output, the output of "sc", and a reproducible test case provides enough output that we'll be able to know what the problem is, (if you want to be extra safe or if the bug is hard to reproduce, include the output of "reg" and "sc"), and often tells us where the problem is and possibly even how to fix it.

The better the test case, (remember: reproducible, simple, clear, and concise), the less you need to worry about the serial debugging output, and the easier it is for us to track down the bugs and fix them.


Developers' Workshop: Square Peg, Round Hole

By Victor Tsou

This week we have the pleasure of fielding a slew of questions regarding BBitmap. These questions come to us courtesy of a little known development house, B&B Technologies, down in Highland, Texas. Lead developer Ms. Lolita grouses:

Q:

Setting pixels in the bitmap is sloooooowwww. How can you guys call yourselves the Media OS? Here's our function for the B_CMAP8 case:

void set_pixel8(BBitmap *bitmap, int x, int y, uchar8 color)
{
  bitmap->SetBits(&color, 1,
    y*bitmap->BytesPerRow()+x, B_CMAP8);
}
A:

SetBits() is a tool intended for importing rather than manipulating images. You were understandably bamboozled by its somewhat misleading name. Setting the color of a pixel is best accomplished with a cocktail composed of a healthy medley of Bits() and BytesPerRow() (mmm...tastier than nachos):

void set_pixel8(BBitmap *bitmap, int x, int y, uint8 color)
{
  *((uint8 *)bitmap->Bits() +
    y*bitmap->BytesPerRow() + x) = color;
}

In other words, think of a BBitmap as nothing more than a pixel buffer. Nothing requires you to draw pixels one at a time; you'll probably find it more convenient and more efficient to work directly with the pointer. For example, setting the buffer to white requires nothing more than:

memset(bitmap->Bits(), 0, bitmap->BitsLength());

Of course, you must always be aware of the color space of the BBitmap when directly altering its contents. SetBits() may prove itself useful by allowing you to normalize the color space to one understood by your rendering methods.

The other engineer on the project, the lovely Ms. Tanqueray, upon observing this interchange, gracefully retorts:

Q:

You mean I have to draw everything myself as I did in the days of yore? What a crock! You should all be fired!

A:

If you're drawing polygons or ellipses or any other primitive found in BView, you always have the option of attaching a BView to the BBitmap and using our preexisting rendering methods. And, of course, you can always manhandle the bits directly even if a BView has been attached to the BBitmap.

After some experimentation, Ms. Tanqueray returns with further BBitmap comments:

Q:

SetBits() is broken. This code is supposed to convert a B_RGB32 bitmap into a B_CMAP8, but all I get is a garbled Technicolor mess!

BBitmap source32(bounds, B_RGB32);
BBitmap dest8(bounds, B_CMAP8);

dest8->SetBits(source32->Bits(),
  source32->BitsLength()/4, 0, B_RGB32);

Would you boys kindly explain what's going on?

A:

Surprisingly, the Be Book provides insight into this problem:

"If the color space mode is B_RGB32, the data should be triplets of three 8-bit components—red, green, and blue, in that order—without an alpha component. Although stored as 32-bit quantities with the components in BGRA order, the input data is only 24 bits in RGB order."

In other words, B_RGB32 indicates data in B_RGB24 order. As a further caution (some might call this a caveat or something), notice how the Be Book craftily avoids any mention of 15-bit, 16-bit, and big-endian color spaces. This is our subtle way of covering up SetBits()'s inability to interpret these spaces as sources or as destinations. Conversions involving these spaces will have to be performed by hand just as in the days of yore.

To summarize:

  1. In most respects, a BBitmap is nothing more than a buffer of pixel data. Setting individual pixels is best performed with the pointer to the buffer returned by Bits(). All you DOS people, think segment 0xA000.

  2. Drawing into a BBitmap may be facilitated by attaching a BView, which in no way hampers your ability to directly manipulate the buffer by hand.

  3. Importing data from an existing source can be handled by SetBits(), but this function is, unfortunately, not as handy as it might seem at first, due to constraints on the allowable color spaces.


The Logic of Linux

By Jean-Louis Gassée

Last week, Netscape made more noise with its endorsement of Linux than it did by posting Navigator source code on the specially created www.mozilla.org site. The official media (The New York Times and others) took notice. They praised the creator, Linus Torvalds, and Linux users who have made Linux into a successful grassroots Unix, with an installed base running into the millions worldwide. Five million is the most commonly used number, but this is difficult to judge, given the product's freeware roots.

Listening to the radio as I drove around the UCSD campus in the rain this past week-end, I noticed that NPR gave prominent play to the event and to the open source software movement. It is nice to see Linux taken seriously by the mainstream media. Perhaps this signals the arrival of Linux on the establishment radar. Seen in this light, if Netscape were to use Linux as their OS platform for enterprise applications, its newfound acceptance would be validated.

Besides its considerable technical virtues, Linux makes much cultural sense for Netscape. The Internet culture is rooted in Unix and in libertarian sharing of software. What better fit for the "liberated" Navigator than an equally unshackled version of Unix? One observer contends the Net runs on free software: sendmail, bind, apache and perl, now we have Linux and Navigator.

In one respect, this results in a re-federation of a very diverse set of people and code around the two related goals of proliferating free software and fighting Microsoft's dominance. Salon has an interesting piece on the topic at www.salonmagazine.com/21st/?st.ne.fd.mnaw. In it, Eric S. Raymond, coauthor of The New Hacker's Dictionary, makes an important distinction between "free" and "open"; while one may not always agree with his perspective, it makes for useful and entertaining reading.

[On a closely related issue, eagle-eyed developers have questioned our adherence to the Open Software pact in relevant parts of our recent release such as the bootloader. Specifically, if we use or modify public domain code, are we putting that source code back into the public domain as we are obligated to do? We're investigating this as I write this and will take whatever corrective action is necessary.]

Back to Netscape, endorsing Linux would complete the manoeuvre started with opening Navigator sources in response to Microsoft's bundling a free browser with Windows. The move accentuates the polarization of the software world. On the right, Microsoft, its proprietary products and its well-publicized business practices, on the left, open software constantly improved by the myriad supporting and sharing programmers.

But, for Netscape, Linux could offer advantages beyond the technical, cultural or political. Consider Netscape's predicament in the enterprise space where Windows NT has been making rather successful inroads. Whenever Netscape succeeds with a product running on Windows NT, there is the risk Microsoft will "climb up their tail pipe," so to speak, by bundling a "free" competitor with the OS as they did with the browser. With Linux, there's nothing of that nature to fear—no one will try and use their control of the OS platform to diminish the prospects of a successful Netscape application.

Is this idle speculation? Perhaps, but note the accent in recent public pronouncements on "open" standards. A little less obvious is the de-emphasis on the "official" Java, because of platform control issues, again. On the Mozilla site, at www.mozilla.org/blue-sky/jvm.html, there is a clear appeal: "It's critical that Mozilla embed a robust, fast, free JVM as quickly as possible. There is no robust, fast, free JVM."

We'd like that too.

Creative Commons License
Legal Notice
This work is licensed under a Creative Commons Attribution-Non commercial-No Derivative Works 3.0 License.