Red Hat revs Enterprise Linux distro
Red Hat today announced the fourth release of its Enterprise Linux commercial distribution, RHEL 5.3, adding support for new hardware, some virtualization tweaks meant to keep pace with recent innovations to hypervisor technology, and all sorts of other goodies.
RHEL 5.3 offers over 150 enhancements. A number of these features made their debut in Fedora, a Red Hat-sponsored project that creates what amounts to a development release for RHEL - although Fedora contributors get annoyed when you say that.
They like to emphasize the freedom that Fedora has as a standalone software package, more or less free from Red Hat strings and encumbrances. But Red Hat uses Fedora as the foundation for RHEL and that, ultimately, is the purpose of the project. Generally speaking, there are tweaks to the Linux 2.6 kernel, device drivers, and various applications that are part of the distro.
First up in RHEL 5.3 is support for Intel's "Nehalem" Core i7 processors, the desktop processors launched last November that are implemented in a 45 nanometer process, that have a new microarchitecture and an Opteron-like interconnect called QuickPath, and that will be delivered in servers (if all goes well) sometime around the end of March.
RHEL 5.3 not only runs on these chips. It also hooks into the chips' power management features (which saves energy), and it's been tuned to take advantage of their simultaneous multithreading. The Xeon variants of Nehalem for servers will scale up to eight cores, with two virtual threads in each core. In a four socket system, that will mean an operating system and its applications will have 64 threads to play with. That's as many threads as the biggest RISC/Unix SMP box offered at the turn of the millennium, and now, you can pack it all onto one motherboard with big gobs of DDR3 main memory.
Interestingly, Red Hat said that when using the desktop versions of the Nehalem chips - which have four cores - commercial applications running atop RHEL 5.3 showed a factor of 1.7 improvement in performance. With number-crunching and bandwidth-eating workloads typically run on workstations and parallel supercomputers, the speed up on Nehalem chips compared is up by a factor of about 3.5.
RHEL 5.3 also includes OpenJDK, an open implementation of Java SE 6 development kit and runtime for Java. (It is based on the same code base as Sun Microsystems' own JDK, according to Red Hat). The company is claiming that the OpenJDK support combined with its JBoss Enterprise Application Platform yields the first enterprise-grade, fully open source Java stack. (There will be plenty of arguing about that point).
RHEL 5.3 includes support for Global File System 2, an update to GFS that was available for evaluation purposes only on RHEL 5.2. The new release also includes support for data encryption for block devices (mainly disk drives) using the Linux Unified Key Setup (LUKS) feature. RHEL 5.3 also has an iSCSI boot firmware table that allows the operating system to be booted from disk arrays linked to servers over an iSCSI link. The ability to have a RHEL server be an iSCSI target (meaning a server can act like a disk array to other servers using iSCSI for connectivity) has moved from a tech preview in RHEL 5.2 to production grade in RHEL 5.3. (Although this feature is also in the tech preview section of the release notes).
The SystemTap system monitoring tool has been "rebased" to level 0.7.2 of that project, which has offers a few new features and some minor enhancements, according to RedHat. These include Cluster Manager, Red Hat Package Manger, Open Fabrics Enterprise Distribution, and OpenSSL. The OFED tweaks are interesting in that they are a set of open source drivers for Ethernet and InfiniBand networks that implement the Remote Direct Memory Architecture, or RDMA, which in turn allows devices on a network to reach directly into the memory of their peers, thus reducing latencies and speeding up software performance.
A RHEL release always has some technology previews, and RHEL 5.3 is no different. These includes the ext4 file system, which is nearly ready for primetime as a kicker to the ReiserFS and ext3 file systems and a bridge to BTRFS (pronounced "ButterFS"), a future Linux file system. The ext4 file system is designed to support volumes up to 1 exabyte in size, and individual files that are as large as 16 terabytes. ext4 will be able to juggle up to 4 billion files. By comparison, the latest ext3 implementations support file sizes between 16 GB and 2 TB, depending on the block size chosen at install time, with volume sizes topping out at between 2 TB and 16 TB.
BTRFS, which might see the light of day in 2010, was launched in June 2007 and is a POSIX-compliant file system. It will support very large volumes (16 exabytes in size) and a large number of files (two to the power of 64 files, to be precise). The file system has object-level mirroring and striping, checksums on data and metadata, online file system check, incremental backup and file system mirroring, subvolumes with their own file system roots, writable snapshots, and index and file packing to conserve space, among many other features under development. Oracle did a lot of the groundwork for the BTRFS project.
The RHEL 5.3 tech previews also include eCryptfs, a stacked cryptographic file system that mounts atop file systems such as ext3, and Stateless Linux, a way to run replicated and read-only Linuxes for distributed workloads. The release also has a tech preview of the GNU GCC 4.3 compiler set (C, C++, and Fortran and their libraries).
Customers who have a support contract will get the updates to RHEL free of charge, since Red Hat does not charge for upgrades on RHEL releases. It won't be long before Oracle and CentOS kick out their clones of RHEL 5.3.
RHEL 5 was launched in March 2007 after about a six-month delay in getting it out the door because of issues concerning the integration of the Xen virtual machine hypervisor with the operating system. Xen was changing so fast then that it was difficult to get RHEL and Xen in synch. This will be less of an issue going forward, now that Red Hat owns Qumranet, which created the alternative KVM hypervisor that is part of the mainstream Linux kernel these days. With RHEL 6, you can bet KVM and not Xen will be the default hypervisor. This is already expected with the Fedora 11 development release due at the end of May, which will be the foundation for RHEL 6 as far as anyone knows. ®
"That's nonsense. There will be no production ready btrfs before 2013. Just look at how long it did take ZFS to be ready (started around 2000)...Soon Linux and Windows are the only major operating systems without ZFS. (Solaris, MacOS X, FreeBSD and NetBSD will all have it)..."
Agreed with everything in your post, bro :)
RE: all the Sunshiners!
Well, so much for Schwartz and his penguin suit, there's obviously still plenty of fear and bitterness in the Sun camp when it comes to RHEL. I'm not surprised - Sun have flip-flopped so many times on Linux the Sunshiners are just confused and afraid.
Look, I know working with Slowaris tends to send you to sleep, but please try and stay awake and pay attention. Yes, we all know Niagara can run 64 weenie threads, the point is the four-socket Nehelem servers will run 64 real threads with a large application base already there for them. This means they will be able to run real enterprise applications like SAP, Oracle or MS SQL with real performance, unlike Niagara which is really only suited to webserving, and unlike Rock which will either be too expensive to compete or just remain vapourware.
"....Soon Linux and Windows are the only major operating systems without ZFS. (Solaris, MacOS X, FreeBSD and NetBSD will all have it)." Ah, you seem to have failed to grasp the concept of "enterprise", which is where the real major OS play and the money is at. Slowaris is barely enterprise anymore, it's losing the high-end share so fast. Mac OS is definately not enterprise, the only place it appears in business is in the crayon department. FreeBSD and NetBSD are both popular with techies but not there at the enterprise level, especially when compared to Red Hat or SuSE. And especially not compared to the real enterprise high-end choices, AIX or hp-ux, both of which (as far as I have been told) have zero plans for ZFS. So there seem to be plenty of "major OS" not interested in ZFS.
And even if ZFS was worth looking at there is still the legal issue of the undecided lawsuit with NetApp and ZFS being a copy of WAFL. Red Hat and SuSE are being smart going with BTRFS over ZFS, because if NetApp win they lose nothing, and if Sun survive long enough to win then Red Hat and SuSE can just integrate ZFS later if it ever gets to the point where it's stable or better than BTRFS. Heck, if Sun win, since they've opensourced it, even hp and IBM can copy it into AIX and hp-ux if they ever get bored of working with Veritas. And Sun make zero return and will carry on losing market share. Face it, you Sunshiners scream and drool about ZFS like it's some magic cure-all that will make everyone love Slowaris and buy Sun, but the truth is it's just another sinkhole for Sun development money with very little return. Meanwhile, the Sun piggybank is getting lighter and lighter.....
Not according to this page http://searchenterpriselinux.techtarget.com/news/article/0,289142,sid39_gci1334319,00.html#
It says 2012, but we all know how those things slip, especially with filesystems.
And regarding ZFS, it's not standing still...