Announcement

Collapse
No announcement yet.

Ubuntu vs. OpenSolaris vs. FreeBSD Benchmarks

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

  • Ubuntu vs. OpenSolaris vs. FreeBSD Benchmarks

    Phoronix: Ubuntu vs. OpenSolaris vs. FreeBSD Benchmarks

    Over the past few weeks we have been providing several in-depth articles looking at the performance of Ubuntu Linux. We had begun by providing Ubuntu 7.04 to 8.10 benchmarks and had found the performance of this popular Linux distribution to become slower with time and that article was followed up with Mac OS X 10.5 vs. Ubuntu 8.10 benchmarks and other articles looking at the state of Ubuntu's performance. In this article, we are now comparing the 64-bit performance of Ubuntu 8.10 against the latest test releases of OpenSolaris 2008.11 and FreeBSD 7.1.

    http://www.phoronix.com/vr.php?view=13149

  • #2
    Conclusions:
    OpenSolaris
    In Java and filesystem related benchmarks, OpenSolaris outperformed the rest. The former due to Java (and Solaris) being owned by Sun and the latter due to OpenSolaris using the ZFS filesystem.

    FreeBSD
    Interestingly, FreeBSD did well (but not as well) in areas where OpenSolaris performed while Ubuntu did very poorly in these areas. FreeBSD is great for ray tracing!

    Ubuntu
    Ubuntu was marginally better at everything else (the majority of benchmarks) that was tested.


    Question: These are all opensource so why isn't someone working at porting all these goodies over so they all work extra fast! I know people are trying to do ZFS. Licensing issues, yada yada. What about this awesome ray tracing power of FreeBSD?

    Comment


    • #3
      Originally posted by Sacha View Post
      Question: These are all opensource so why isn't someone working at porting all these goodies over so they all work extra fast! I know people are trying to do ZFS. Licensing issues, yada yada. What about this awesome ray tracing power of FreeBSD?
      FreeBSD doesn't have a built in ray-tracer, it's a matter of scheduling and stuff, and that's a very complex issue; at best we can learn from how BSD works, Linus isn't going to core a scheduler that's been nearing perfection for years just because of a benchmark.


      I'd be fascinated to see Snow Leopard thrown in here once it comes out.

      Comment


      • #4
        @Sacha

        One test won't tell you if system is better to do something then other.

        ---

        @Michael

        It looks you are deaf to people suggestions. I can understand that you wanted to make this benchmark compatible with previous one, but why not make one test extra with bonnie++ and larger file? Following to results of your benchmarks it looks like Mac OS beats other systems which are known of their very high performance. You know, it's really funny :> Even more funny are people who take those benchmarks serious.

        Probably that is correct use of bonnie++ benchmark:

        http://home.comcast.net/~jpiszcz/benchmark/allfs.html
        Last edited by kraftman; 11-24-2008, 05:19 PM.

        Comment


        • #5
          With Java and OpenSolaris being products of Sun Microsystems, it shouldn't be too surprising that the fastest Java performance was generally witnessed atop this Sun operating system.
          Well, that could be the reason, but I suspect the fact that the Solaris system was running update 10 of Java 6 has something to do with this as well. Update 10 introduced a LOT of improvements over previous updates, better performance might be one of them.

          Comment


          • #6
            it would nice to see a linux that is not cursed with ext3

            btw, if you care about filesystem performance, read this;
            http://bulk.fefe.de/lk2006/bench.html

            in short, solaris lies, bsd sucks.

            Comment


            • #7
              Michael,

              Thanks for an exciting roundup!

              Originally posted by kraftman View Post

              The ZFS in the bonnie++ benchmark here may explain the reason why Solaris shined! It has very low cpu demands in many of the tests. Both Solaris and MacOS use ZFS.

              Unfortunately, Linux cannot readily use ZFS bcause of its license, CDDL (http://en.wikipedia.org/wiki/CDDL):

              In the words of Danese Cooper, who is no longer with Sun, one of the reasons for basing the CDDL on the Mozilla license was that the Mozilla license is GPL-incompatible. Cooper stated, at the 6th annual Debian conference, that the engineers who had written the Solaris kernel requested that the license of OpenSolaris be GPL-incompatible. "Mozilla was selected partially because it is GPL incompatible. That was part of the design when they released OpenSolaris. [...] the engineers who wrote Solaris [...] had some biases about how it should be released, and you have to respect that".
              However, there is the FUSE project http://zfs-on-fuse.blogspot.com/ , which makes Linux run ZFS!

              If this would be included in future tests one may see the importance of file system as such.

              Comment


              • #8
                Originally posted by sabriah View Post
                Michael,

                The ZFS in the bonnie++ benchmark here may explain the reason why Solaris shined! It has very low cpu demands in many of the tests. Both Solaris and MacOS use ZFS.
                Leopard doesn't use ZFS. It uses something which is just piece of crap - Drag wonderfully explained that in another thread. As far as I know Apple plans to use ZFS in future versions of Mac OS. There's Linux file system which uses very low CPU too - JFS, but its performance is probably lower than EXT3 etc.
                Last edited by kraftman; 11-24-2008, 05:13 PM.

                Comment


                • #9
                  Originally posted by kraftman View Post
                  Leopard doesn't use ZFS. It uses something which is just piece of crap - Drag wonderfully explained that in another thread. As far as I know Apple plans to use ZFS in future versions of Mac OS. There's Linux file system which uses very low CPU too - JFS, but its performance is probably lower than EXT3 etc.
                  Leopard can use ZFS as a read only option right now. Snow Leopard will introduce R/W capability. Drag's arguments also were extremely outdated (some just plain out wrong). Many of his concerns were addressed in later versions of HFS+.

                  Comment


                  • #10
                    idea: all this tests with virtualized hosts too

                    Originally posted by phoronix View Post
                    Phoronix: Ubuntu vs. OpenSolaris vs. FreeBSD Benchmarks
                    virtualisation is coming.

                    it would be great to see a test of this 3
                    OS on virtualized hardware.

                    for example via KVM.
                    of course with the same virtual hardware and
                    the same amount of ram for every OS.

                    Comment


                    • #11
                      Need more information about the environment.

                      Originally posted by phoronix View Post
                      Phoronix: Ubuntu vs. OpenSolaris vs. FreeBSD Benchmarks

                      Over the past few weeks we have been providing several in-depth articles looking at the performance of Ubuntu Linux. We had begun by providing Ubuntu 7.04 to 8.10 benchmarks and had found the performance of this popular Linux distribution to become slower with time and that article was followed up with Mac OS X 10.5 vs. Ubuntu 8.10 benchmarks and other articles looking at the state of Ubuntu's performance. In this article, we are now comparing the 64-bit performance of Ubuntu 8.10 against the latest test releases of OpenSolaris 2008.11 and FreeBSD 7.1.

                      http://www.phoronix.com/vr.php?view=13149
                      The introduction said you left things in their default state which might not be the best thing for a beta of FreeBSD. There is a strong possibility that it has debug options on which could impact performance.

                      I'm installing BETA2 currently (there's an RC2 out as well). So I'll post an update soon with the generic kernel configuration

                      Comment


                      • #12
                        Originally posted by sbryant View Post
                        The introduction said you left things in their default state which might not be the best thing for a beta of FreeBSD. There is a strong possibility that it has debug options on which could impact performance.

                        I'm installing BETA2 currently (there's an RC2 out as well). So I'll post an update soon with the generic kernel configuration
                        There's no doubt about it: FreeBSD runs with debugging turned on in all betas, and it's turned off in the RCs. It's certainly not uncommon to have debugging turned on in operating systems during the beta phases. I'm not sure why he chose 7.1beta2 instead of 7.0 release.

                        Comment


                        • #13
                          Originally posted by rivald View Post
                          There's no doubt about it: FreeBSD runs with debugging turned on in all betas, and it's turned off in the RCs. It's certainly not uncommon to have debugging turned on in operating systems during the beta phases. I'm not sure why he chose 7.1beta2 instead of 7.0 release.
                          Maybe because of scheduler? I don't think so if final release will perform noticeable better then beta. Btw. using default settings in all systems is stupid in my opinion. I bet that tweaked Linux kernel can perform ways better.
                          Last edited by kraftman; 11-25-2008, 03:26 PM.

                          Comment


                          • #14
                            Originally posted by kraftman View Post
                            Maybe because of scheduler? I don't think so if final release will perform noticeable better then beta. Btw. using default settings in all systems is stupid in my opinion. I bet that tweaked Linux kernel can perform ways better.
                            There are many things that affect performance, especially if there's more stuff to wade through. Think about it, allocating and setting memory to a certain value takes way from doing other things. Despite the scheduler, all that does is making sure it stays responsive, nothing can be done about the amount of time things require to complete. If it just allocates trash data, that's a lot less time then allocate, write.

                            On top of that if you have checks in the code for helping debug or generating a sensible core then ofcourse it's going to take up time. You should read up on what these debug options do.


                            Debug options (the big two are WITNESS and INVARIANTS):
                            http://www.freebsd.org/doc/en/books/...g-options.html

                            Comment


                            • #15
                              Should be noted that some distro's also run their Alpha's and Beta's and RC's as well with debugging turned on. It's not limited to BSD.

                              Comment

                              Working...
                              X