Page 3 of 5 FirstFirst 12345 LastLast
Results 21 to 30 of 50

Thread: Minix 3.2 Released, Uses LLVM/Clang, SMP, ELF

  1. #21
    Join Date
    Jun 2011
    Posts
    827

    Default

    Quote Originally Posted by Nobu View Post
    Actually, rather than "how much slower", I'd be more interested in "how much more stable". Has anyone tried writing (for instance) a real network driver that causes havoc on purpose?
    Off hand, i can't point you to a link. But this kind of stuff has definitely been done. i recall watching a talk (most likely on youtube) with Tannenbuam discussing this very idea (of using drivers to cause havoc on purpose, in order to test the microkernel). If i remember correctly, they never were able to make the microkernel/system crash, only the driver, itself. That is one of the points of microkernels, so naturally, it would only be logical to do real-world testing of the design.

    cheerz

  2. #22

    Default

    Quote Originally Posted by ninez View Post
    If i remember correctly, they never were able to make the microkernel/system crash, only the driver, itself. That is one of the points of microkernels, so naturally, it would only be logical to do real-world testing of the design.

    cheerz
    What's the point of microkernel when your file system crashes?

  3. #23
    Join Date
    Oct 2010
    Posts
    413

    Default

    Well, if your filesystem crashes, I'd imagine there's still the possibility to debug and figure out why it crashed without having to reboot, assuming the kernel is still alive and you have the necessary utilities in memory or accessible via some other means.

  4. #24
    Join Date
    Nov 2011
    Posts
    267

    Default

    Quote Originally Posted by Nobu View Post
    Well, if your filesystem crashes, I'd imagine there's still the possibility to debug and figure out why it crashed without having to reboot, assuming the kernel is still alive and you have the necessary utilities in memory or accessible via some other means.
    Which is what kdb does for Linux.

  5. #25
    Join Date
    Jun 2011
    Posts
    827

    Default

    Quote Originally Posted by kraftman View Post
    What's the point of microkernel when your file system crashes?
    Ideally, the file-system restarts after it crashes ~ without any significant impact on the rest of the system. ie: it doesn't crash. The whole idea of a Microkernel/Self-healing OS is that all of it's components are isolated from each other and use IPC to communicate. So if a driver / file-system / *insert component here* happens to crash - it won't take down the whole operating system. ie: it is self-healing.

    If you're interested in the subject, rather than asking in the Phoronix forums - just search around the web && watch a video or two on youtube. there are lots of videos, whitepapers/research papers, wiki's, etc.

    But here are a couple quick links...Tannanbaum, discussing Minix3;

    http://www.youtube.com/v/bx3KuE7UjGA?

    and

    http://en.wikipedia.org/wiki/Minix

    QNX is another Self-healing/Microkernel/OS that is used in various industries.

    http://www.qnx.com/products/neutrino...rino-rtos.html

    and..

    http://en.wikipedia.org/wiki/QNX

    I am pretty sure that QNX is in much wider use than Minix (probably ever will be), and has been for years (dating back to the late 80s). It is used for many industrial applications. Also, (not that i am a huge fan of Blackberry). but as of 2012, all of Blackberries new smartphones will be using QNX, and currently, the blackberry Playbook does use QNX ---> actually let me rephrase that, they are using a modified version of QNX called BBX (BackBerry + QNX).

    cheerz
    Last edited by ninez; 03-01-2012 at 03:11 PM.

  6. #26

    Default

    Quote Originally Posted by ninez View Post
    Ideally, the file-system restarts after it crashes ~ without any significant impact on the rest of the system. ie: it doesn't crash. The whole idea of a Microkernel/Self-healing OS is that all of it's components are isolated from each other and use IPC to communicate. So if a driver / file-system / *insert component here* happens to crash - it won't take down the whole operating system. ie: it is self-healing.
    I'm just interested in data safety. I don't care if my system won't crash while my data can be screwed up same time (so the need for the restart or not won't make any difference then...) I see no reason to use micro kernel. Not only it's many times slower, but if it can't protect my data there's simply no real advantage. I don't like listening Tanenbaum, because the guy is not sane.

  7. #27
    Join Date
    Jul 2008
    Location
    Greece
    Posts
    3,776

    Default

    Quote Originally Posted by kraftman View Post
    I'm just interested in data safety. I don't care if my system won't crash while my data can be screwed up same time (so the need for the restart or not won't make any difference then...) I see no reason to use micro kernel. Not only it's many times slower, but if it can't protect my data there's simply no real advantage. I don't like listening Tanenbaum, because the guy is not sane.
    He is actually very sane. One of the top in the field.

    And who told you that microkernels will trash your data? And why would you assume that? That doesn't even make sense.

  8. #28
    Join Date
    Jun 2011
    Posts
    827

    Default

    Quote Originally Posted by kraftman View Post
    I'm just interested in data safety. I don't care if my system won't crash while my data can be screwed up same time (so the need for the restart or not won't make any difference then...) I see no reason to use micro kernel. Not only it's many times slower, but if it can't protect my data there's simply no real advantage. I don't like listening Tanenbaum, because the guy is not sane.
    I have to agree with RealNC, here. (no both accounts) Tannanbaum is quite, sane and does make some valid points. For some types of applications one actually might want to have a self-healing OS, that never crashes and performance may not be a huge deal. And also remember, depending on the design - the performance loss may actually be fairly marginal. (like 5% or something). I don't think data loss is a real problem, either. ~ as i've never read anywhere that data loss was a problem for these kinds of systems.

    Going back to Blackberry using the BBX microkernel ~ My cousin is in the hospital right now, and has a PlayBook. So i was playing with it the other week, while visitng him... i didn't find the PlayBook to be slow, or bogged down because it uses a Microkernel. I don't have a link, but recently i saw Blackberry's next-gen tablet demoed, and it was actually quite fast and smooth (the video was 10minutes long, with a lot of navigating, running android applications, using the web and watching videos).

    so i imagine, at least with QNX - the performance loss is nothing to complain about.
    Last edited by ninez; 03-01-2012 at 07:02 PM.

  9. #29
    Join Date
    Feb 2012
    Location
    Kingston, Jamaica
    Posts
    291

    Default Torvald's minions

    It would serve Torvalds' minions to read a little about microkernels and play with a few without dismissing them just because Torvalds said so. Update your knowledge of micrkernels from the state it was in the 1990s.

    Mach and some older microkernels performed poorly. Minix in its previous incarnations may have performed badly. That doesn't mean that microkernels are inherently bad. In fact, several commercial and research microkernels offer acceptable performs (worse case: about 10% performance loss.). I am talking about kernels like QNX and various L4 kernels (OKL4, Pistachio, Fiasco(.OC) etc...).

    If raw speed is the only thing that monolithic kernels have to offer over microkernels then I would prefer have a microkernel in my gadgets because stability is more important to me.

    In every microkernel/monolithic kernel I've come a cross, I've always seen the question "But what if your disk crashes?....". If a filesystem crashing is the only argument you have against microkernels then you are missing the point and you are just stupid and ugly ((c) Torvalds). Disk drivers aren't the only drivers that may crash. If I'm around my desktop toiling away all day, not interested in listening to music then my sound driver gives out, I would certainly not want to see a kernel panic. A disk device driver may be fatal or not to the system in a microkernel, fair enough, but microkernel proponents do not claim that microkernels guarantee 100% stability. Rather, it increases stability, so the number of misbehaving drivers may move from a total of 8 (arbitrary number) down to a grand total of 1. I would rather to have 1 landmine on my lawn than 8.

    I haven't read in detail but minix is supposed to be able to recover from a disk driver crash (take with salt till I can confirm).

    All in all, you guys bashing microkernel and Tanenbaum should be more open minded and try things before putting them down. You sound like the kind of people who would have considered Noyce crazy back in the day.

    Microkernels you can look on incude minix, l4 from tu dresen, l4 from Karlsuhe, l4 from nicta & oklabs, l4 from blabs, helenos and see what they have to offer.

  10. #30

    Default

    Quote Originally Posted by RealNC View Post
    He is actually very sane. One of the top in the field.
    Ignoring his envy regarding Linux and reality that has already proven Linux model is better... I doubt.

    And who told you that microkernels will trash your data? And why would you assume that? That doesn't even make sense.
    Nobody, because I'm not saying microkernel will trash my data. It's obvious bugs happen everywhere. If there's a bug in the file system microkernel won't help you. Thus, I don't care if I'll have to restart my box which runs sometimes hundreds times faster or just restart my file system while in both cases my data will be lost.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •