Slashdot Mirror


NYT Ponders the Future of Solaris In a Linux/Windows World

JerkBoB links to a story at the New York Times about the future prospects of Sun's Solaris, excerpting: "Linux is enjoying growth, with a contingent of devotees too large to be called a cult following at this point. Solaris, meanwhile, has thrived as a longstanding, primary Unix platform geared to enterprises. But with Linux the object of all the buzz in the industry, can Sun's rival Solaris Unix OS hang on, or is it destined to be displaced by Linux altogether?"

6 of 340 comments (clear)

  1. Re:Not all the best features are technical by the+eric+conspiracy · · Score: 4, Informative

    Solaris doesn't play in the supercomputer market because the Sparc architecture is not cost effective for that application. The big iron that Solaris runs on are enterprise scale database servers, which are optimized for an entirely different set of performance parameters.

  2. Re:Performance by Anonymous Coward · · Score: 5, Informative

    Got the balls to drop an strace on your production Oracle database? I tried strace on an Oracle database on RHEL 5 and the damn process deadlocked and the box needed a reboot to clear it up. Good thing it was a development DB.

    I've put a truss (and now dtrace) on PRODUCTION Oracle databases running on Solaris many times.

    I don't dare do that on Linux.

    Solaris is as far beyond Linux in stability as Linux is beyond Win2K.

  3. Re:PC-BSD anyone? by Anonymous Coward · · Score: 3, Informative

    Actually there are quite a few features found in Linux not found in FreeBSD or PC-BSD. Like write access to NTFS,

    Write access to NTFS is not likely to be an issue on a server.

    inotify

    FreeBSD has kqueue, which is arguably far better designed and generically useful an event notification mechanism

    many journalled file systems,

    Why do you need more than one journaled file system?

    user-space FS

    FreeBSD supports FUSE. I'm not sure why you want this on a server, though

    multiple kernel virtualization methods and client optimizations

    You got one right! OMG! Linux supports KVM, FreeBSD doesn't include any built-in virtualization support. But then, FreeBSD has jails, and Linux doesn't, so we'll call it a draw?

    stable LVM, et al

    FreeBSD has GEOM, a "modular disk I/O request transformation framework", which is arguably more generic and better designed than LVM. Notice a trend here ...

    And how about those forklift upgrades that FreeBSD seems to love... no "apt-get dist-upgrade" foolishness for BSD. Oh no... it's time to format!

    FreeBSD has freebsd-update, which supports binary updates between major and minor releases. Prior to the introduction of freebsd-update, the process was a bit more painful, but a format was certainly not required

    However, note that FreeBSD doesn't tie their third party software (eg, packages/ports) to the base system. So you can keep a FreeBSD 4.x system around for 5 years and still have a modern up-to-date installation of your applications.

    On Debian, however, you'll just have Stale-OS, unless you're willing to risk destabalizing your production systems by upgrading the *entire* base operating system

    You're clearly unfamiliar with FreeBSD

  4. Re:Performance by chekk4 · · Score: 3, Informative

    Actually, ZFS really IS that impressive. Almost platter speed using the equivalent of raid-6 without a hardware raid controller. It's not perfect, but in my testing, it has performed extremely well.
    How do you get 1.4TB/s with only 138 drives? Even if you have hundreds of gigabit switches, the drives collectively are nowhere near that aggregate throughput.

    Cheers.

  5. Re:Virtualization makes Solaris less relevant by C0vardeAn0nim0 · · Score: 4, Informative

    just atended a seminar about solaris LDOMs at the office. with LDOMs (which only run on niagara T1 and T2 CPUS) you can assign a whole PCI bus to a single guest OS in such way that the guest has full control of that bus, which implies no performance penalty if that bus have a couple of HBAs to connect it to a storage/SAN.

    the sun guy at the seminar (same instructor that ran the solaris 10 administration course I took last year) made it very clear that the kind of hardware backed virtualization provided by LDOMs is the second best for high I/O apps, losing only for hardware partitioning, and it's way ahead of the kind of full software virtualization provided by vmware, virtual pc or virtual box.

    --
    What ? Me, worry ?
  6. Re:Or else... by Saint+Aardvark · · Score: 5, Informative

    Blastwave...heh. Which Blastwave are you talking about?

    Sorry, this is a bit of a sore point for me. At work, we have a Solaris 10 machine that powers about 30 SunRays for mathematicians. JDS is fine, but adding other programs is a pain. (Disclaimer coming up, so bear with me.)

    • Blastwave: They just had the split. But before that there were problems. Upgrading CUPS broke printing; they'd moved around some Ghostscript filters. Upgrading Postfix broke Postfix, because they'd moved the config files to play nicer with zones, and their script that should've dropped everything in the right place didn't. These were stable versions, not the unstable.
    • pkg-src: Great until you trip over something that won't compile and spend days trying to track down what it is -- say, 1 package in 20. Sounds like good odds? Try compiling Firefox or Kile, with dependencies stretching back to libc and the Dead Sea scrolls. I'm guessing they just aren't able to do as much testing on Solaris...and fair enough; the job of making umpty thousand packages compile on mumble different OS' is hard enough.
    • compile from source: fine, unless it's obscure (say, some mathematical package) that assumes GNU tools all the way, or a Linux OS, and weird, obscure things break.
    • download binaries: yes, if they've got 'em.

    And now for the disclaimers: No, this isn't enterprise (which was your point; I was looking for a place to jump into this discussion, and the mention of Blastwave got me). Yes, a real sysadmin could compile all this from scratch without problem. Yes, this is an edge case on top of an edge case (desktops for mathematicians? How obscure!). Yes, ZFS and dtrace are seriously, jaw-droppingly awesome.

    But this is my experience; so far, I simply have not done anything remotely enterprise. It's all been server + desktop in small shops. And for that environment, requirements are changing all the time. The mail server now needs to do spam filtering and DNS. Yes, they should be split up, but there isn't the budget. The new guy wants KDE on his machine instead of Gnome, or needs to try out a new library to see if it works.

    And for these, it's not "set it and forget it"; we need new packages, or updates to the old ones, all the time. If all the heartache I described was a one-time thing, I'd do it and be done...but in this environment, it'll need to be done again in three months. That means a good package manager (hello, Debian!), or a good ports tree (*BSD), or an environment that everyone is familiar with (Linux, because it has just that much mindshare).

    Bit of a rant, and less coherent than I'd like. But it's 6am, I haven't had my coffee yet, and my kid's about to wake up...so I'll have to leave it there.