Announcement

Collapse
No announcement yet.

OpenSolaris vs. Linux Kernel Benchmarks

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by drag View Post
    I really really dislike HFS+ file system.

    You know that HFS+ isn't even POSIX compliant? It doesn't even use the same symantics for locating data...
    [snip]
    Or something like that. They use the BSD-VFS layer in the BSD-half of the XNU hybrid kernel to make it look and act like a POSIX-ish file system.

    HFS+ is also NOT a journaling file system! It's a file system much more like VFAT.. in fact it is from the same generation of file systems. HFS+ pre-dates NTFS, Ext3, and all that stuff. The journaling features are another VFS add-on.. and they actually have quite a bit more overhead then actually having journaling in the file system.

    This is true from the OS X 10.2-10.3 days (which after that I stopped caring about OS X), they may have fixed it... but that isn't likely. It's not a fun file system to deal with...
    For every time I had a file system related issue on OSX/HFS+ I wish I had a nickel ! It is not only an ancient, non POSIX compliant file system it also is the most unreliable FS used on modern OSes. Both NTFS and ext3 are far more reliable. And no doubt ZFS takes it even further.

    I am not a big fan of OSX when it comes to Operating System Design - I hate several things - HFS+ is one (why do I need permissions database repair in 2009), Running 32-bit kernel on 64-bit processor while supporting 64-bit binaries (I heard this is getting addressed in next release), threading fiasco (funnels) to name a few. Just look at how many months they are taking to integrate ZFS - not even RO support is mainstream yet - that speaks volumes about design and scalability.

    Given that I was very surprised to see OSX winning benchmarks against Linux and Solaris - but as you implied and I hinted in the earlier post, the benchmarks in question certainly do not seem to be representative of real-world performance.

    Comment


    • #17
      Originally posted by drag View Post
      I really really dislike HFS+ file system.

      You know that HFS+ isn't even POSIX compliant? It doesn't even use the same symantics for locating data...

      Unix:
      /path/to/directory/filename.txt

      HFS+
      volumename:to:directory:filename.txt

      Or something like that. They use the BSD-VFS layer in the BSD-half of the XNU hybrid kernel to make it look and act like a POSIX-ish file system.

      HFS+ is also NOT a journaling file system! It's a file system much more like VFAT.. in fact it is from the same generation of file systems. HFS+ pre-dates NTFS, Ext3, and all that stuff. The journaling features are another VFS add-on.. and they actually have quite a bit more overhead then actually having journaling in the file system.

      This is true from the OS X 10.2-10.3 days (which after that I stopped caring about OS X), they may have fixed it... but that isn't likely. It's not a fun file system to deal with...
      Did you know that there have been many changes since you last looked at OS X? Did you know that it is 100% posix compliant and certified as of 10.5?

      Also HFS+ does not predate NTFS, NTFS debuted in 1993, HFS+ didn't debut until 1998. As far as overhead goes there is very little overhead at all for journalling. Not even appreciablely measured. Don't confuse HFS+ with HFS which is an entirely different filesystem. Some people seem to think that HFS+ is just HFS with more capabilites added to it because they can be read by HFS only machines which would be completely wrong. That is done through a wrapper found on HFS+ to allow HFS capable machines to read them instead of going "huh the disk is empty".

      You can as well very easily use the same syntax for paths as well. I really have no idea where you got that. (Actually I do, that would be from HFS on "classic" systems").

      As far as reliabilty goes I ran OS X servers for quite a few years with an average traffic rate of 40TB / month per server, not once has data been corrupted or lost spanning years.

      Last edited by deanjo; 05-14-2009, 08:21 PM.

      Comment


      • #18
        Originally posted by deanjo View Post
        Did you know that there have been many changes since you last looked at OS X? Did you know that it is 100% posix compliant and certified as of 10.5?
        Isn't there a difference between UNIX 03 compliance and POSIX compliance?

        Comment


        • #19
          Originally posted by parapup View Post
          Isn't there a difference between UNIX 03 compliance and POSIX compliance?
          Same thing.

          IEEE Std 1003.1 aka ISO/IEC 9945

          http://www.unix.org/version3/
          "Read/Download IEEE Std 1003.1, 2004 Edition, Single UNIX Specification Version 3"

          http://posixcertified.ieee.org/
          Last edited by deanjo; 05-14-2009, 09:10 PM.

          Comment


          • #20
            Originally posted by deanjo View Post
            Did you know that there have been many changes since you last looked at OS X? Did you know that it is 100% posix compliant and certified as of 10.5?
            So?

            I was talking about the File system and not the operating system.

            This is why OS X supplied UFS. UFS is indeed a POSIX file system and is provided for compatibility reasons.

            The reason why HFS+ is used by default is becuase the GUI, the non-UNIX stuff, needs to have it's resource forks for them to work properly and you'd run into problems on UFS.


            Also HFS+ does not predate NTFS, NTFS debuted in 1993, HFS+ didn't debut until 1998.
            Ok.

            As far as overhead goes there is very little overhead at all for journalling.
            Enabling journalling on a older PowerMAC is a good way to have a massive drop in performance. They could probably of optimized it quite a bit since it was originally introduced with the 10.2.2 OS (which was not enabled by default), but HFS+ is still relying on a VFS layer for it's journalling features and is not something that is built into the actual FS.


            Beleive me. I know this because I was trying to deal with managing ~90 Mac OS X workstations when that was new and was getting really tired of having to rescue their file systems and repair permissions.


            Not even appreciablely measured. Don't confuse HFS+ with HFS which is an entirely different filesystem.
            I wasn't. HFS+ has been around since MacOS 8.1.

            I just was mistaken since NTFS is such a obviously superior FS that I assumed it was introduced after HFS+ was.



            You can as well very easily use the same syntax for paths as well. I really have no idea where you got that. (Actually I do, that would be from HFS on "classic" systems").
            No.

            HFS+ uses : as a path deliminator and all paths must be full paths.

            Like I said before the BSD-VFS layer is used to trick the Unix half of the operating system into thinking it's on a POSIX-like file system.

            I ran into these issues quite often when I was trying to combine my Unix scripting skills with Apple's GUI stuff to automate things for students. If you think that openning up a terminal gives you access to what is happenning 'underneath' that pretty GUI that OS X uses... your going to be dissappointed.


            As far as reliabilty goes I ran OS X servers for quite a few years with an average traffic rate of 40TB / month per server, not once has data been corrupted or lost spanning years.
            Ya... a server system with a reliable, UPS-backed. power supply is not a situation were you would tend to run into problems.

            I helped administrate the Macs used in the electronic imaging and graphics classrooms of 2 college campuses. Total number of Macs were about 200+. Which is going to be about the largest concentration of Macs that your likely to see anywere around were I live.

            At the time they ranged from beige tower G3s to the dual proccessor G5 machines. The most common being the PowerMac G4s.

            They are a lot funner to work with then the Windows workstations that were used for 3D graphics.. but to say that I was less then impressed by the robustness of the OS.

            The fact that Apple is still using HFS+ means that not really that much has changed.

            If Apple was to move to ZFS (as the default for everything) then that would be very very very good move and very impressive.

            However I am convinced that Apple doesn't give a flying fart about the server market or anything to do with the 'enterprise' weither it was in desktops or server deployments. Home users couldn't give a shit less about the file system and there really isn't any good reason for them to do so.. just as long as they use Time Machine with external media effectively, which is exactly why Apple can get away with having a lousy file system.
            Last edited by drag; 05-15-2009, 04:41 AM.

            Comment


            • #21
              Originally posted by drag View Post
              So?

              I was talking about the File system and not the operating system.

              This is why OS X supplied UFS. UFS is indeed a POSIX file system and is provided for compatibility reasons.

              The reason why HFS+ is used by default is becuase the GUI, the non-UNIX stuff, needs to have it's resource forks for them to work properly and you'd run into problems on UFS.
              Errr, wrong, UFS was dropped as an installable option in 10.5. Not needed anymore.





              Beleive me. I know this because I was trying to deal with managing ~90 Mac OS X workstations when that was new and was getting really tired of having to rescue their file systems and repair permissions.
              Wow a whole 90 systems, that would cover the employee lounge systems on the size of OS X networks I've worked on. The OS X networks I've managed range from 400-700 seats and users were given full admin privileges as they were to needed to fully replicate issues of Mac users. Their workstations were their test machines.

              No.

              HFS+ uses : as a path deliminator and all paths must be full paths.

              Like I said before the BSD-VFS layer is used to trick the Unix half of the operating system into thinking it's on a POSIX-like file system.

              I ran into these issues quite often when I was trying to combine my Unix scripting skills with Apple's GUI stuff to automate things for students. If you think that openning up a terminal gives you access to what is happenning 'underneath' that pretty GUI that OS X uses... your going to be dissappointed.
              You really haven't used OS X in a while, scripting in OS X is no different then any other unix when it comes to syntax. The path delimiters are all handled at a kernel level, transparent to the end user completely.

              Ya... a server system with a reliable, UPS-backed. power supply is not a situation were you would tend to run into problems.

              I helped administrate the Macs used in the electronic imaging and graphics classrooms of 2 college campuses. Total number of Macs were about 200+. Which is going to be about the largest concentration of Macs that your likely to see anywere around were I live.

              At the time they ranged from beige tower G3s to the dual proccessor G5 machines. The most common being the PowerMac G4s.
              The networks I managed were far greater in size and in a hostile environment. Workstations had no UPS back up and still data loss was not an issue on those at all either. Those ranged from the very first iMacs running 9 and 10.0 all the way up to 10.5 and on the latest hardware. People would do a hardshut downs and power would go out all the time. Still data loss was not present. If there was data loss it was because of hardware failure.

              If Apple was to move to ZFS (as the default for everything) then that would be very very very good move and very impressive.
              Sure ZFS would be nice, it would be nice on all OS's.

              However I am convinced that Apple doesn't give a flying fart about the server market or anything to do with the 'enterprise' weither it was in desktops or server deployments. Home users couldn't give a shit less about the file system and there really isn't any good reason for them to do so.. just as long as they use Time Machine with external media effectively, which is exactly why Apple can get away with having a lousy file system.
              lol Time machine was not intended to make up for filesytems, it's there for when users have a brain fart and delete something they wanted back later. Your right though, Apple concentrates on it's desktops market but to say they don't give a shit about their enterprise users is completely false. They wouldn't go through the bother of writing custom formware for their xserve's ADMs if that was the case. They would do like so many other venders and use plain jane consumer drives that have firmware that is optimized for windows use. You can read more about this and the other testing practises here http://db.tidbits.com/article/10166.

              Comment


              • #22
                Why not take latest development edition for openSolaris? snv 112 included in Solaris express edition, for example.

                Comment


                • #23
                  You really haven't used OS X in a while, scripting in OS X is no different then any other unix when it comes to syntax. The path delimiters are all handled at a kernel level, transparent to the end user completely.
                  Well good.

                  Because when trying to combine Apple Scripting features for the GUI with Unix script for munging files and mounting file systems it was a huge pain in the ass to be translating between the HFS+ paths for the Apple scripting into Unix paths for the shell scripting.

                  Errr, wrong, UFS was dropped as an installable option in 10.5. Not needed anymore.
                  Well I suppose that means that you can no longer install OS X to UFS.. which is fine. I'd never want to install OS X on UFS anyways. The non-Unix half of OS X never worked well at all and was never a recommended configuration.

                  However that does not mean that I would actually try to use HFS+ for anything serious. Housing the OS is one thing, but actually using it as a active storage volume is quite another.

                  I am not going to trust HFS+ with anything important on the Unix side of the fence.

                  Even with all the work Apple has put into it HFS+ and their Posix file system emulation stuff can't handle even something as simple and fundamental as multiple hard links correctly. If you try to create multiple hard links to a file on HFS+ Apple creates a wacky little very-hidden file in those directories as part of their work around for not supporting it. Trying to use multiple hardlinks is prone to all sorts of corruption and nasty corner cases.

                  I know you keep thinking I am making this stuff up. But remember you were the one originally telling me that : was only used on the ancient HFS stuff and that HFS+ uses Unix paths.

                  Seriously. How can anybody think that HFS is POSIX-compliant or a safe Unix FS to use when it doesn't even support something as fundamental as case sentitivy? POSIX says that file names are to be treated as nothing more then a string of bytes. That's it. The only illegal characters are 'null' and '/'... everyhting else goes. Capital letters, lower case leters, control characters, newlines, tabs, spaces, etc etc. All these things are legal and are used to resolve individual file names.

                  Sure case insentivity is nice to have since it is much more user friendly by far.. but it's definately not POSIX compatible. POSIX compatibility is completely overrated if your looking at desktop usability... but the high level of complexity and just plain weirdness of HFS+ is not good and means that it's error prone no matter what.

                  If you don't believe me...
                  http://rixstep.com/2/2/20070718,00.shtml
                  Last edited by drag; 05-15-2009, 06:15 PM.

                  Comment


                  • #24
                    Originally posted by drag View Post
                    W POSIX says that file names are to be treated as nothing more then a string of bytes. That's it. The only illegal characters are 'null' and '/'... everyhting else goes. Capital letters, lower case leters, control characters, newlines, tabs, spaces, etc etc. All these things are legal and are used to resolve individual file names.

                    Sure case insentivity is nice to have since it is much more user friendly by far.. but it's definately not POSIX compatible. POSIX compatibility is completely overrated if your looking at desktop usability... but the high level of complexity and just plain weirdness of HFS+ is not good and means that it's error prone no matter what.

                    If you don't believe me...
                    http://rixstep.com/2/2/20070718,00.shtml
                    Case sensitivity has been present in HFS+ for a long time now. You have options of either case sensitive or non case sensitive since 10.3. Also if you would read your own link you would be see that it's referring to and written pre Leopard. Again, old and outdated info.
                    Last edited by deanjo; 05-15-2009, 06:39 PM.

                    Comment


                    • #25
                      Originally posted by L33F3R View Post
                      Sun (now oracle) has its own breed of CPU.
                      (Sun is not "now Oracle"... until Oracle sign on the dotted line, which is expected to happen in a few months time, it's business as usual as separate companies.)

                      Comment


                      • #26
                        I question the validity of the SQL benchmark results. Can you explain to me how the Linux system is supposed to be able to handle thousands of transactions per second, while being ACID compliant? That's just not physically possible with a hard drive. All this benchmark tells me is that the storage subsystem of Linux is in a bad shape, if it allows things like this to happen (and if it's not a database configuration issue).

                        Comment


                        • #27
                          Originally posted by bugmenot View Post
                          I question the validity of the SQL benchmark results. Can you explain to me how the Linux system is supposed to be able to handle thousands of transactions per second, while being ACID compliant? That's just not physically possible with a hard drive. All this benchmark tells me is that the storage subsystem of Linux is in a bad shape, if it allows things like this to happen (and if it's not a database configuration issue).
                          Is this real world benchmark? Nope. In real world benchmarks results will be much different and then you'll see real Linux advantage over os x (or over Solaris which isn't performance monster there). It's, because of database configuration, but I'm not sure if it's an issue, because I don't know what was tested.
                          Last edited by kraftman; 06-11-2009, 01:28 PM.

                          Comment


                          • #28
                            OpenSolaris vs Linux Kernel Benchmarks

                            As with the Linux distros, you should chose one binary repository and stick with it from the start, I mean load ON Operating system and Network then chose one only repo.

                            Also, GNUish last OpenSolaris has its own package management, so, do not install legacy binary providers, or you get into trouble.

                            Also be certain to keep in mind that OpenSolaris has become a generic name for all non pure Sun Solaris products.
                            Indiana .NOT. Milax.NOT. Solaris Express.NOT. Nexenta OS.NOT. SchilliX.NOT. marTux.NOT.Solaris
                            Analogy with Linux distros, all Linux kernel based, you cannot mix repos from Ubuntu with NovellSUSE or Mandriva and RHat.

                            Comment

                            Working...
                            X