Red Hat Bashes Sun As It Launches Enterprise Linux 4

by Timothy Prickett Morgan

With the launch of Enterprise Linux 4 at the LinuxWorld expo in Boston, commercial Linux distributor Red Hat demonstrated that is has grown sufficiently as a commercial entity. So much so that its marketing spin now sounds rough and almost right. While Red Hat was keen to demonstrate that Version 4 of its Linux was a suitable and better replacement for Sun Microsystems' Solaris Unix variant, RHEL 4 really is a Linux that is a viable alternative to any Unix, and more significantly, other Linux 2.6 variants.

This message was somehow lost in the Sun bashing that Red Hat is increasingly engaged in, which is in part a response to the Red Hat bashing that Sun's top brass has been engaging in since they decided that taking Solaris 10 open source and making it compete head-to-head, heart-to-heart, and soul-to-soul with the commercialized versions of Linux--of which Red Hat is the dominant commercializer--was the best long-term strategy for Sun. Hammering on Sun is a business that IBM and Hewlett-Packard have certainly engaged in since 2001, and they have had a reasonable amount of success. But what Red Hat doesn't seem to understand (or at least acknowledge) is that the Solaris base is considerably larger than the Red Hat installed base. Sun reckons that it has shipped over 750,000 betas of Solaris 10, and about 4 million commercial Solaris 8 and Solaris 9 licenses to date. The total worldwide installed base of Linux servers is around 2 million, and Sun probably has a few million machines running earlier Solaris 2.5 and 2.6 editions. If Red Hat does really well in calendar 2005, cumulative sales (since the company was founded) might reach 750,000 Red Hat Enterprise Linux licenses., A very large portion of these (probably 50 percent) will be renewals from prior years' sales, making Red Hat's installed base of Linux instances somewhere around 300,000 to 400,000. (That's very rough, back-of-the-envelope math based on the limited information Red Hat gives about its installed base.)

Sun has certainly had a hard four years, making it a supposedly easy target to pick on. But Solaris on Sparc has about 12,000 applications, and the porting job from Solaris on Sparc to Solaris on X86 is a lot easier than moving from Solaris to Red Hat Linux, which had just over 1,000 enterprise-class applications that have been certified by some 300 software developers as of July 2004. And Sun's open source (and free) OpenSolaris variant, its wickedly inexpensive new support options for Solaris 10, and its embracing of cheap X86 iron makes Solaris 10 absolutely credible and affordable compared to Red Hat Enterprise Linux 4 and Novell SUSE Linux Enterprise Server 9. To my eye, it looks like Red Hat wrote its marketing campaign over a year ago and hasn't realized the Solaris landscape has changed.

That said, the fact that the Red Hat Network, Red Hat's automated customer support infrastructure will be able to "manage" Solaris servers and workstations and their applications in addition to Red Hat Linux systems and applications will probably be appealing to some joint shops or those who are indeed in the process of moving from Solaris to Linux. The Red Hat Network Management Module for Solaris is expected by the end of the second quarter of this year. No word on what it might cost.

If Red Hat were serious about taking on Sun, it would pull a page from the Sun playbook and add native support for Solaris applications inside a runtime environment inside the future Red Hat Enterprise Linux 5, and it would have announced its intention to do so at LinuxWorld last week. Sun's Solaris 10 has a feature called Project Janus, which is a runtime environment that can run compiled Red Hat applications inside a shell that translates Linux calls to Solaris calls on the fly. Janus is not, strictly speaking, an emulation environment. aIt is an API translation layer with relatively low overhead. Sun intends to add support for other Linuxes in Project Janus as well. With Solaris 10 now being open sourced (and the full code base for Solaris expected to be delivered soon), Red Hat could do a turnabout and pull the same trick on Sun. Red Hat missed a true zinger of an opportunity by not seeing it could do this. (Red Hat has similar compatibility libraries in RHEL 4 that allow it to run code for its 2.1 and 3 versions.)

Red Hat officials say hundreds of applications are in the process of being certified on Enterprise Linux 4 and more than 750 servers and workstations have been certified as of the launch with one of the four variants of the operating system. As we have previously reported, Enterprise Linux 4 and Red Hat Desktop, the cut-down version for PCs and workstations, are all based on the Linux 2.6.9 kernel. Click here to see a RHEL 4 salient characteristics table I built to save you from having to read a lot of text describing the editions.

RHEL 4 has a stunning number of new features that make it an absolutely credible enterprise platform, some of which come from the open source community's development of the Linux 2.6 kernel and some of which have been created by Red Hat for Enterprise Linux 4 and which are only now being released into the community. (The GNU General Public License says new ideas and code must be shared, but it doesn't say do it before releasing it to your own customers first.) The fact that Red Hat has four different versions that span from a desktop PC to a 16-socket servers and that it has not raised its license prices after a pretty substantial price hike last year means Red Hat has a pretty good chance at increasing its installed base. But it has a long, long way to go before it unseats that vast installed Unix base out there.

While many of the features in the Linux 2.5 development release and the Linux 2.6 commercial release were backported by Red Hat to Enterprise Linux 3, some features, such as the block I/O layer from Linux 2.6, which significantly improves server performance, could not be ported to the Linux 2.4 kernel at the heart of RHEL 3. The new Linux supports different I/O operation schedulers, each of which is optimized for different workloads. The NOOP scheduler requests I/O tasks exclusively, and it assumes that Linux will be running in a virtual environment where that virtual partition on a server will determine the best order for I/O operations. RHEL 4 also has a completely fair queuing (CFQ) round-robin I/O scheduler, a deadline scheduler (which provides better access to I/O for high I/O applications like databases), and an anticipatory scheduler, which uses heuristic analysis and the fact that I/Os tend to be synchronous and sequential while writes tend to be asynchronous and random to give a 1 millisecond delay to look for a sequential read. By doing so, for the cost of 1 millisecond, you can save anywhere from 10 to 16 milliseconds if you hold a read until after the sequential read is truly done.

RHEL 4 supports Serial ATA disk drives as well as the ATA, SCSI, and Fibre Channel disks Red Hat has been supporting for years; the software also includes an improved ext3 file system that allows multithreaded applications to do reads and writes from four to five times faster when many threads are running. The ext3 file system can now scale to 8 TB. The software also includes the Logical Volume Manager 2 (LVM2), a significant update of the open source LVM1 software that Red Hat and SUSE shipped with their Linux 2.4 releases. Red Hat says LVM2 will soon have a new multipathing I/O driver as well as support for RAID1 mirroring of volumes; these features are not yet in RHEL 4.

Security is a big deal with Linux 2.6, and it is one of the main factors driving the option of Linux over Windows. Enterprise Linux 4 includes the Security Enhanced Linux (SELinux) security extensions developed by the U.S. National Security Agency available as an option, allowing Red Hat to push Linux into sensitive areas where high security is of paramount importance. SELinux has a policy set for every application that described every kernel function it is allowed to access. The effect of these policies is that an application can be compromised by a hacker, but that compromised application can be shut down at the kernel level and it cannot reach over into other applications to corrupt them. SELinux creates a containment field around all applications.

Red Hat and Novell have both committed to get their respective Linux 2.6 variants certified at the Common Criteria EAL4+ level as soon as possible, giving their Linuxes the same or higher certification as the popular Unixes out there today. Further on the security front, RHEL 4 supports Intel's Execute Disable (XD) and AMD's No Execute (NX) processor security features. Red Hat has also added checks in the glibc memory allocator to do sanity checks so it can detect the double freeing of memory and heap buffer overflows, two common techniques for hacking systems. Red Hat has also added gcc compiler buffer bound checking and changes to the printf format string that prevent other common exploits. These last tweaks are not yet in other commercial Linuxes.