Linux 2.4 contains several improvements. For desktop use, the most notable improvement is support for a multitude of new devices, including those that plug into the universal serial bus port. For server use, the most significant changes are improvements that will let Linux take better advantage of systems with multiple processors.

Additional Platform support

Linux now runs on the following platforms:

• IA64
• IBM S/390
• SuperH
• Newer 64-bit MIPS

Linux Internals

Linux 2.2 was a major improvement over Linux 2.0 and the Linux 1.x series. It supported many new filesystems, a new system of file caching, and it was much more scalable. (If you want a list of features new to Linux 2.2, you can read my article about it.) Linux 2.4 builds on these things and more to be the best darned Linux kernel yet in a variety of situations.

The Linux kernel is an assortment of modular components and subsystems including device drivers, protocols, and other component types. APIs, programming interfaces that provide a standard method by which the Linux kernel can be expanded, glues these to the core of the Linux kernel. Most of this document will focus on these components of the Linux OS as these are the components that seem to do the most work: drive your disks, read your files, and do all of the obvious and physical things. Linux 2.4 is however much more than just these components. These assorted drivers and APIs all revolve around a common center of the Linux kernel. This center includes such fundamental features as the scheduler, the memory manager, the virtual filesystem, and the resource allocator.

Linux 2.4 is the first release of the Linux kernel that will include a full-featured resource management subsystem. Previous incarnations of Linux included some vestiges of support, but it was considered kludgy and did not provide the functionality needed for the "Plug and Play" world that we live in today. Unlike many of the other internal changes, many users will be able to directly experience this change as it impacts the way resources are allocated and reported in the kernel. As part of this change, the PCI card database deprecated in Linux 2.2 has been un-deprecated so that all resources can have an associated device name, rather than just an associated driver.

The new release of the Linux kernel also fixes some problems with the way the VFS (virtual filesystem) layer and the file caches were handled. In previous versions of Linux, file caching was dependent on a dual-buffer system, which simplified some issues, but caused many headaches for kernel developers who had to make sure that it was not possible for these buffers to be out of synch. Additionally, the presence of the redundant buffer increased memory use and slowed down the system as the kernel would have to do extra work to keep things in synch. Linux 2.4 solves these problems by moving to a simpler single-buffer system.

A number of changes in Linux 2.4 can be described as "enterprise level." That is, they may not be immediately useful to many desktop users but work to strengthen Linux as a whole. For the most part, the addition of these features does not degrade Linux in more "normal" environments. First, Linux 2.4 can handle many more simultaneous processes by being more scalable on multiprocessor systems and also by providing a configurable process limit. Second, the scheduler has been revised somewhat to be more efficient on systems with a larger number of concurrent processes. Third, the revised Linux kernel can now handle an amazing number of users and groups-- about 4.2 billion. (And that's a lot of users!) In addition, support for more powerful hardware is provided in the new kernel, which now supports up to 64 gigabytes of RAM on Intel hardware, up to 16 Ethernet cards, 10 IDE controllers, multiple IO-APICs, and other pointless abuses of good hardware. The 2-gigabyte file size restriction has also been lifted. With these changes and others, the Linux kernel development team is proving that Linux can be an option in many new environments.

The way Linux handles shared memory has also been changed in Linux 2.4 to be more standards compliant. One side effect of this set of changes is that Linux 2.4 will require a special "shared memory" filesystem to be mounted in order for (POSIX-style) shared memory segments to work. SysV-style shared memory will continue to work without the additional filesystem. The mounting of this filesystem should be handled by the distributions when they become ready for Linux 2.4.

Another "odd" change included in this new revision of the Linux kernel involves the VFS layer. In previous versions of Linux, indeed most every version of UNIX, you can only mount a filesystem once. Linux 2.4 has decided to turn this on its ear by allowing any filesystem to be mounted as many times as desired (with all changes appearing immediately on the other mount points). This is especially useful for filesystems like /proc that need to be mounted in an initrd and then mounted again. Additionally, the infrastructure is there to eventually allow even odder things: union mounts which contain files from multiple filesystems together, etc. In this respect, Linux is either cutting edge or crazy. I'm betting on the former. :)

Linux 2.4 also includes a much larger assortment of device drivers and supported hardware than any other Linux revision and any particular device you care to name has a decent shot at working under Linux 2.4. (Of course, you should consult the documentation before you go out and buy any new hardware, just in case. New hardware especially may not be supported yet.)

One frequently asked question about Linux 2.4 is how much memory it will require. Many operating systems seem to require more and more memory and resources as they mature, but Linux 2.4 will largely buck that trend by actually requiring less memory in certain situations. Of course, Linux 2.4 includes much more functionality than does Linux 2.2and many of these features do take up space so your mileage may vary. (Remember that most kernel components can be disabled at compile-time, unlike the bloat of many other operating systems.)

UNIX Devices - /dev/*

Although taking place largely under the hood, there is one major change that will have a drastic effect on the way Linux 2.4 handles devices. Incidentally, it is also one of the new kernel's most debated features: DevFS, the Device Filesystem. Although currently optional, when enabled the new Device Filesystem will have two very major effects on the way users and programs interact with hardware and other devices. First and most obvious, nearly every "standard" device name has changed. Instead of putting all devices essentially in the root directory of the /dev tree, Linux 2.4's /dev tree leans towards having a large number of directories corresponding to controllers or general groupings with individual disks and other devices in those directories. For example, what was "/dev/hda" is now likely to appear somewhere under "/dev/ide0" This change dramatically increases the available namespace for devices and allows for easier integration of USB and similar device systems. The second major effect of the new system is that device nodes will no longer be created by a user and stored on a local disk. Instead, device nodes are created by the drivers that use them, as they are loaded. This not only makes the /dev directory cleaner, but it also allows Linux to use a root filesystem that may not understand all UNIX semantics (such as a DOS filesystem.) An additional userspace daemon, devfsd, will be recommended so that the old style names will continue to work.

For the most part, this change should only affect bleeding edge users and distribution maintainers. (It remains to be seen what work distributions decide to do in this area before release.) Unfortunately, there are some drawbacks to this system for power users. For example, it is more difficult in this new system to custom name your devices or assign non-standard permissions to them. It is commonly believed however that these issues can be solved using user-level scripting.

System Buses - ISA, PCI, USB, MCA, etc

Processors are just one small part of the nifty world that exists inside your computer. Equally important is the computer's bus architecture, the component(s) of the system that is often responsible for internal and external devices. Some bus architectures, such as the original ISA, are more irresponsible towards their hardware than anything else-- they don't provide any resource management functionality,just a place to put in cards. Others, such as PCI, support much more advanced levels of configuration and allow for devices to be relocated in memory and other things. As of Linux 2.2, all major buses used on Intel hardware are supported including (E)ISA, VLB, PCI, and the newest addition(an older bus only popular on IBM hardware), MCA. Linux 2.4 expands on this support by including direct support for ISA Plug-and-Play devices (a scheme to make ISA devices almost as intelligent as PCI ones) and I2O devices. But perhaps most importantly, Linux 2.4 is the first version of the Linux kernel to provide a robust system for resource management. During the development of Linux 2.4, it became apparent that such a system would be a requirement if Linux were to ever completely support USB, PC cards, or any number of other modern hardware advances and take its rightful place as a "modern" operating system.

ISAPnP has long been a major issue for Linux users. Although support for ISA hardware has dwindled in favor of more robust PCI hardware, many budget devices are still sold that use ISAPnP. Previously, Linux users could get ISAPnP hardware working by using often frustrating pnp utilities that could require hours of tweaking to get quite right. Some distributions attempted to automate this process, but none met with any great level of success. As the Linux 2.4 development progressed, it became apparent that it would be easy and beneficial to integrate ISAPnP support into the resource manager.(Although as of this writing, not all drivers have been recoded to take advantage of its features.) Sadly however, Linux 2.4's support for ISAPnP at the kernel level comes at a time when actual PnP hardware is relatively uncommon Had this functionality been available earlier, more users could have benefited.

In contrast, Linux is right on the bleeding edge with its support for I2O devices, a new more "intelligent" superset of PCI. Relatively revolutionary in its day, PCI was a great improvement as it provided for centralized management of devices' memory ranges, registers, etc. I2O devices go the next level by providing an API at the device level that will allow OS independent drivers to be provided for devices. The underlying OS then need only understand the "generic" I2O APIs to use the device instead of the hardware-specific ones. As this technology is relatively new, not many devices have been manufactured to take advantage of this yet, but Linux will be ready if and when they show up in the marketplace.

Much of the major work with devices recently has not been with the internal busses, but rather with external ones such as the PC Card bus, and the various serial busses. The most common variety of external device is the PC Card (formerly known as a PCMCIA card). Linux 2.4 includes, at long last, support for these devices in a stock kernel. (Previously, it was possible to download a driver from an external source; nearly every distribution actually chose to do this.) Of course, an external daemon and other components will still be required to make the most out of these devices.

Perhaps the most exciting news on this front is the Universal Serial Bus (USB), an external bus that is coming into prominence for devices such as keyboards, mice, sound systems, scanners, and printers. USB is a popular option on many new pieces of hardware, including non-Intel hardware. Linux's support for these devices is still in early stages but a large percentage of common USB hardware (including keyboards, mice, speakers, etc.) is already supported in the kernel.

And even more recently, Firewire (IEE1394) support has been added into the Linux kernel. Firewire is a popular option for many high-bandwidth devices. Not many drivers (or devices) exist for this hardware architecture yet, but this support is likely to improve over time, as the architecture matures.

Block Devices - LVM, Disk Drives, etc

In its simplest form, a block device is a device, which can be expressed as an array of bytes that can be accessed non-sequentially. This would include devices such as disks (where you can read any sector you want) but not serial ports (because you can only read what is at the end of the wire.) Extensions to this concept (such as ejecting a disk, etc.) are handled in Linux through ioctls (I/O Controls). The concept of block devices hasn't changed in quite a while and support for things such as IDE and SCSI disk drives has been present since the first revisions of the Linux kernel.

In Linux 2.4, all the block device drivers have been rewritten somewhat as the block device API has been changed to remove legacy garbage from the interface and to completely separate the block API from the file API at the kernel level. The changes required for this API rewrite have not been major. However, module maintainers who have modules outside the main tree may need to update their code. (One should never assume full API compatibility for kernel modules between major revisions.)

On the desktop at least, disk devices that use the IDE bus are most prevalent. Linux has supported IDE since the earliest kernels, but Linux 2.4 has improved support for these devices in a number of ways. First off, high-end systems that have multiple IDE controllers may benefit as the number of supported IDE controllers has been increased from 4 to 10. As most motherboards are shipped with a maximum of two, this is not likely to impact many desktop users. Secondly, there have been changes in the IDE driver, which will improve Linux 2.4's support for PCI and PnP IDE controllers, IDE floppies and tapes, DVDs and CD-ROM changers. And finally, Linux 2.4 includes driver updates, which should work around bugs present in some IDE chip sets and better support the advanced features of others, such as ATA66.

While it would seem that the SCSI subsystem has not changed as much as the IDE subsystem, the SCSI subsystem has been largely rewritten "under the hood." Additionally, a number of new SCSI controllers are supported in this release. A further SCSI cleanup is expected sometime during the2.5 development cycle.

One completely new feature in the Linux 2.4 kernel is the implementation of a "raw" I/O device. A raw device is one whose accesses are not handled through the caching layer, instead going right to the low-level device itself. A raw device could be used in cases where a sophisticated application wants complete control over how it does data caching and the expense of the usual cache is not wanted. Alternatively, a raw device could be used in data critical situations where we want to ensure that the data gets written to the disk immediately so that, in the event of a system failure, minimum data will be lost. Previous incarnations of this support were not fit for inclusion as they required literally doubling the number of device nodes so that every block device would also have a raw device node.(This is the implementation that many commercial UNIXes use.) The current implementation uses a pool of device nodes which can be associated with arbitrary block devices.

One huge area of improvement for Linux 2.4 has been the inclusion of the LVM (Logical Volume Manager) subsystem into the mainstream kernel. This is a system, standard in Enterprise-class UNIXes such as HP-UXandTru64 UNIX (formerly Digital UNIX), that completely rethinks the way filesystems and volumes are managed. Without going into detail, the LVM allows filesystems to span disks, be resized, and be managed in a more flexible way. Some of the features of the LVM subsystem can be replicated with the md (multiple device) driver or some userspace tools. However, the LVM subsystem offers this support in a (de facto) standards-compliant manner that will also be at least somewhat familiar to users of commercial UNIXes.

Another huge area of improvement for Linux 2.4 will be in its support for RAID devices, multiple disks working together to provide redundant storage or speedier read accesses. In the new kernel, nearly the entire RAID subsystem has been rewritten. Performance, probably the most important facet of a complete RAID implementation, has been improved on both SMP and uniprocessor systems. (On SMP, it's now better threaded.)Additionally, the code has been made much more robust with the ability to make RAID arrays recursive and be able to mount sets for a root disk without the use of a ramdisk image. As more enterprise-level users approach Linux, features like a robust RAID subsystem become gating factors between acceptance and non-acceptance. Linux 2.4 again raises the bar.

Partition Tables

Linux has often been considered the Rosetta stone of operating systems. It should not some as a surprise then that Linux 2.4 will include much improved support for "foreign" partition tables on devices. Linux 2.2 and earlier editions of Linux allowed, for example, Macintosh partition tables to be read in the PPC or m68k ports, as they would be most likely to come across them. On Intel hardware, it was possible for the kernel to understand the "standard" IBM format, BSD disk labels, and some other common extensions. In Linux 2.4 however, the walls between ports have come tumbling down and it will now be possible to seamlessly use an external Macintosh harddrive attached to an Intel Linux box. Essentially, Linux will be able to read any partition table format it understands on any port -- if the drivers are loaded in at compile time.

Expanded File System Support

• ReiserFS File System

  Reiser FS is a journalling fulesystem for the Linux Operaing System. This file system can be compiled as a module.

• UDF File System

  The UDF File System is the common filesystem used on DVD disks. This file system can be compiled as a module.

• EFS

  This file system allows you to read and write SGI partitions. This file system can be compiled as a module.

• Journaling Flash Filesystem

  A journalling file system used in embedded devices.

• RamFS

  An in-memory filesystem used in embedded devices that is capable of resizing itself on the fly by being smart about the VFS layer.

• HPFS Improvements

  Added write capability to the OS/2 filesystem.

• NextStep UFS

  NextStep users will be able to mount their CD-ROMs under Linux as we now support an extension to the UFS filesystem that NextStep uses.

Character Devices

The class of devices that can only be accessed sequentially is the character device. These are devices, such as serial devices, which allow you to read from a stream or push data onto it, but not to "skip" ahead or behind. This includes serial and parallel ports, keyboards, mice, and terminal devices. There have been several major improvements in this area for the latest incarnation of the Linux kernel.

One of the largest improvements in this area is in regards to Linux 2.4's support for keyboards and mice. Previous incarnations of Linux included support for serial and PS/2 mice and keyboards (and ADB, for instance, on the Macintosh.) Linux 2.4 also supports using keyboards and mice attached to the USB ports. Additionally, Linux 2.4 also supports keyboards on some systems where the keyboard is not initialized by the BIOS and systems that have trouble determining whether a keyboard is attached or not. And finally, Linux 2.4 includes expanded support for digitizer pads and features an emulation option to allow them to be used as normal mice, even when this is not directly supported in hardware.

Linux's support for serial ports has not changed much since the days of Linux 2.2. Linux 2.4 (and some later versions of Linux 2.2) supports sharing IRQs on PCI serial boards; previously, this feature was limited to ISA and on-board serial ports. Additionally, Linux 2.4 has added a number of new drivers for multi-port serial cards. It is hoped that these changes and others will make using your serial ports under Linux2.4 easier than before.

In a separate department, there has been some work since 2.2 on supporting so-called "WinModems" (or "soft modems"). These are modems which exist largely in software and whose drivers are often only provided by the manufacturer for Windows. (Often the DSP or other parts of the hardware must be implemented in software rather than on the board.) While no code has been submitted to Linus for the support of these beasts, several independent driver projects have been working to get some support for these beasts in and the first fruits of these labors are becoming usable outside the main tree. While it will be a long time before we see most of these devices supported under Linux, for the first time it actually appears that the Open Source snowball is beginning to roll in this direction.

Linux 2.4 also includes a largely rewritten parallel port subsystem. One of the major changes in this area is support for so-called "generic" parallel devices. Programs which access the parallel ports in unusual ways or, more likely, just want to probe the port for PnP information can use this functionality. Additionally, this rewrite allows Linux 2.4 users to access all the enhanced modes of their parallel ports, including using UDMA (for faster I/O) if supported by the hardware. Under the new Linux kernel, it is also possible to direct all console messages to a parallel port device such as a printer. This allows Linux to match the functionality of many commercial UNIXes by being able to put kernel and debug messages on a line printer.

Multimedia: Sound, TV, Radio, etc.

On the complicated side of the character device list, we have some of the less essential devices to be supported by Linux. Linux, in its emerging role as a desktop platform, tries very hard to support sound cards, TV and radio tuners, and other sound and video output devices. To be honest, Linux 2.4 does not include as many groundbreaking changes as Linux 2.2 did in this respect. Linux 2.4 does, however, include updates and new drivers for a variety of sound and video cards, including adding full duplex support. Linux 2.4 and some later versions of Linux 2.2 also include code that will allow some sound devices to more easily allocate memory in required ranges; this should make the configuration and use of some cards much easier.

Work is in progress on a completely rewritten sound subsystem, which will support many of the more advanced features of today's sound cards. This support will not be present in Linux 2.4, but may make it into the kernel for Linux 2.6.

Video Cards

Another more complicated variety of device is the frame-buffer, a way of looking at many video cards. A frame-buffer is simply a section of memory that represents (or is) video memory to such an extent that writing to this memory affects the colors of the pixels on a screen. This is more complicated than most other devices because it requires ioctls to change the palette and to perform other video-related functions.

Linux 2.4 includes a number of new drivers and improvements to old drivers. Especially important here is Linux's support for many more "standard" VGA cards and configurations, at least in some modes. (Even if the mode is only 16 colors-- at least it works.) Please remember that this feature can be bypassed and (on i386) is only necessary for people with certain systems, which cannot be supported in any other way. At this time,the XFree project provides many more drivers to many more video cards than the kernel can support so it is not necessary to use this feature to get support for the X Window System. (SVGAlib and other libraries also allow you to do direct video manipulation on supported hardware, however the use of these libraries must be done carefully as there are some security concerns and race conditions.)

One of the biggest changes in this respect is the addition of the Direct Rendering Manager to the Linux kernel. The DRM cleans up access to the graphics hardware and eliminates many ways in which multiple processes that write to your video cards at once could cause a crash. This should improve stability in many situations. The DRM also works as an entry point for DMA accesses for video cards. In total, these changes will allow Linux 2.4 (in conjunction with Xfree4.x and other compatible programs) to be more stable and more secure when doing some types of graphics-intensive work. These changes should also make some kinds of television tuner cards more workable under Linux.

Networking Protocols

Networking and network hardware is one of the major areas where Linux has always excelled. These devices are neither "character" nor "block" but inhabit a special space free of the need for device nodes. Linux 2.4 will include many improvements to this layer including new drivers, bug fixes, and new functionality added on to existing drivers.

The Linux model of network sockets is one, which is standard across most UNIX variants. Unfortunately however, the standard does have some correctable deficiencies. Under Linux 2.2 and previous versions, if you have a number of processes all waiting on an event from a network socket (a web server, for instance), they will all be woken up when activity is detected. So, for every web page request received, Linux would wake up a number of processes, which would each, try and get at the request. As it does not make sense for multiple processes to serve the same request, only one will get to the data; the remainder will discover nothing to process and fall back asleep. Linux is quite efficient at making this all happen as quickly as possible, however it could still be made more efficient by removing the redundant wakeups. Linux 2.4 includes changes, which implement "wake one" under Linux that will allow us to completely remove this "stampede effect" of multiple processes. In short, "wake one" does exactly as its name indicates: wakes up only one process in the case of activity. This will allow applications such as Apache to be even more efficient and make Linux an even better choice as a web server.

Linux 2.4 also includes a completely rewritten networking layer. In fact, it has been made as unserialized as possible so that it will scale far better than any previous version of Linux. Additionally, the entire subsystem has been redesigned to be as stable as possible on multiprocessor systems and many possible crashes have been eliminated. In addition, it contains optimizations to allow it to work with the particular quirks of the networking stacks in use in many common operating systems, including Windows. It should also be mentioned at this point that Linux is still the only operating system completely compatible with the letter of the IPv4 specification (Yes, IPv4; the one we've been using all this time) and Linux 2.4 boasts an IPv4 implementation that is much more scalable than its predecessor.

As part of this major rewrite, the firewall and IP masquerading functionality of the kernel has been completely rewritten again. (Older users may remember that these same components were largely rewritten for Linux 2.2 also.) The new subsystem has been split into two parts: a packet filtering layer and a network address translation (NAT) layer. These new subsystems are considerably more generic than their predecessors, and it is now possible to do most types of sophisticated (level 3) routing through any Linux box. Previously, this kind of functionality was largely only available with dedicated and proprietary routing hardware. Unfortunately, this major rewrite also includes yet another new userspace tool to manage the available functionality. For compatibility, modules exist which will allow you to use either the Linux 2.0 (ipfwadm) or Linux 2.2 (ipchains) tools without a major loss of functionality. This will make the upgrade from either of these kernel versions relatively seamless.

One major new extension that has now been added to Linux's networking stack is ECN, Explicit Congestion Notification. In a nutshell, ENC allows compliant routers to notify a Linux box that a route is congested. Linux will then respond by reducing the speed at which packets are being sent. In the long run, this will allow Linux boxes to drop less packets over the congested routes and not spend as much time or bandwidth with dropped packets and retransmits.

For Enterprise-level users, there are a number of features that will better enable Linux to integrate into older and newer components of existing network infrastructures. One important addition in this respect is Linux 2.4's new (partial) support for the DECNet and ARCNet protocols and hardware. This allows for better interoperation with specialized systems, including older Digital/Compaq ones. Also of special interest to this class of users, Linux 2.4 will include support for ATM network adapters for high-speed networking.

For the low-end desktop users, PPP is an important part of day-to-day life. Linux 2.4 includes some major rewrites and modularization of much of the code, including a long awaited combination of the PPP layers from the ISDN layer and the serial device PPP layer, such as for dial-up connections with modems. In addition to the modularity, ISDN has been updated to support many new cards. The PLIP (PPP over parallel ports) layer has also been improved and uses the new parallel port abstraction layer. And finally, PPP over Ethernet (PPPoE, used by some DSL providers) support has been added to the kernel.

Although not present in Linux 2.4, there is work now on supporting the NetBEUI protocol used by MS operating systems. While Microsoft will be moving away from this protocol in its products and towards TCP/IP, this protocol is still important for a number of Windows-based network environments.(Previously, kernel developers had commented that the protocol is too convoluted and buggy to be supported in the kernel. Now that an implementation has surfaced, it remains to be seen whether it will be stable enough to ever be in an official kernel.)

PCMCIA (PC Card)

PCMCIA (PC Card) support is now partially implemented in the main kernel source.

Expanded Documentation

Linux now includes a very complete Documentation directory that contains documentation on the various features and capabilities of Linux. These Web Pages were mainly contructed from these files.

Miscellaneous Changes

• Kernel Web Daemon - a kernel module that can process HTTP requests without having to communicate with any user-space programs.
• Accessibility - Linux 2.4 supports speech synthesizer cards which allow vision-impaired Linux users to hear all Linux output, including kernel messages very early in the boot process.