Announcement

Collapse
No announcement yet.

OpenSolaris vs. Linux Kernel Benchmarks

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

  • OpenSolaris vs. Linux Kernel Benchmarks

    Phoronix: OpenSolaris vs. Linux Kernel Benchmarks

    Earlier this week we delivered benchmarks of Ubuntu 9.04 versus Mac OS X 10.5.6 and found that the Leopard operating system had performed better than the Jaunty Jackalope in a majority of the tests, at least when it came to Ubuntu 32-bit. We are back with more operating system benchmarks today, but this time we are comparing the performance of the Linux and Sun OpenSolaris kernels. We had used the Nexenta Core Platform 2 operating system that combines the OpenSolaris kernel with a GNU/Ubuntu user-land to that of the same Ubuntu package set but with the Linux kernel. Testing was done with both 32-bit and 64-bit Ubuntu server installations.

    http://www.phoronix.com/vr.php?view=13826

  • #2
    The OpenSolaris kernel is 32 bit or 64 bit ??

    Comment


    • #3
      Originally posted by iznogood View Post
      The OpenSolaris kernel is 32 bit or 64 bit ??
      Depends on CPU - AFAIK it automatically boots 64-bit kernel on a capable CPU, falling back to 32-bit when CPU doesn't support it. So if Michael used a recent CPU (it is likely he did) it should have been a 64-bit kernel.

      Comment


      • #4
        So we are saying Mac OS X > Linux > Solaris here? If you ask me that sounds weird. I would prefer running Solaris in a hostile production environment and would venture into running Linux only after exercising caution but I would never want to run OSX even on moderately busy production server no matter what.

        Even ignoring my personal beliefs and going with the mainstream ones, it sounds like Micahel should stick in a disclaimer of sorts just to save the innocent some pain (One would feel tempted to buy a Xserver and run their production server on OSX after reading the results of these benchmarks! )

        But anyway I do think these benchmarks serve a good purpose - no harm in making Linux faster in all benchmarks, relevant or not!

        Comment


        • #5
          Originally posted by parapup View Post
          Depends on CPU - AFAIK it automatically boots 64-bit kernel on a capable CPU, falling back to 32-bit when CPU doesn't support it. So if Michael used a recent CPU (it is likely he did) it should have been a 64-bit kernel.
          So since he used a Core duo 2, 32 bit Linux had a huge disadvantage. Hopefully he can do the test again with 64 bit Jaunty and ext4. IMO that would be more accurate
          Last edited by iznogood; 05-14-2009, 09:41 AM.

          Comment


          • #6
            Originally posted by iznogood View Post
            So since he used a Core duo 2, 32 bit Linux had a huge disadvantage. Hopefully he can do the test again with 64 bit Jaunty and ext4. IMO that would be more accurate
            But he included results for 64-bit Hardy with 2.6.30 kernel in the article. The only significant disadvantage to Linux IMHO is not having ext4 when OpenSolaris was using ZFS.

            Comment


            • #7
              Originally posted by parapup View Post
              So we are saying Mac OS X > Linux > Solaris here? If you ask me that sounds weird.
              It sounds weird, because it wasn't real-world benchmark. Compare those results to previous Linux vs Open Solaris test. There wasn't huge difference between them. Also, if you look at previous kernel benchmarks it will probably be even more weird for you. As someone said:

              Reality has no place here


              I would prefer running Solaris in a hostile production environment and would venture into running Linux only after exercising caution...
              Can you explain? I have very different opinion.
              Last edited by kraftman; 05-14-2009, 10:45 AM.

              Comment


              • #8
                solaris runs on X86 hardware but this is not intended for a server environment by any means . Sun (now oracle) has its own breed of CPU.

                Comment


                • #9
                  What happened to the graphs on half the results? Seem to be missing important data, e.g. the SQLite one only has figure for the Solaris test, Ubuntu ones are all empty!

                  You're missing data on:

                  Threaded Data 64Mb Write - 4 Threads : Ubuntu amd64, 2.6.30
                  SQLite : All the Ubuntu ones.
                  PostgreSQL: Solaris
                  Last edited by Garp; 05-14-2009, 11:33 AM.

                  Comment


                  • #10
                    Originally posted by kraftman View Post
                    It sounds weird, because it wasn't real-world benchmark.
                    Right - that's what I wanted to point out, although I was deliberately obtuse.

                    Originally posted by kraftman View Post
                    Can you explain? I have very different opinion.
                    Saner defaults, lesser need to update, better behavior under insane load, unreasonable expectations of backward compatibility etc. - mostly the type of things which you don't want to deal with. In a very chaotic, shared environment where it proves to be a better guard against stupid, careless people. That plus KSSL - transparent crypto acceleration on T2 boxes is a feature that is very useful for providing security to applications that either don't understand crypto or where there is loss of performance doing excessive amount of SSL (JVM 1.4 for e.g.).


                    I am sure most of those things could be dealt with when going with Linux but people just don't want to care anymore

                    Comment


                    • #11
                      Originally posted by Garp View Post
                      What happened to the graphs on half the results? Seem to be missing important data, e.g. the SQLite one only has figure for the Solaris test, Ubuntu ones are all empty!

                      You're missing data on:

                      Threaded Data 64Mb Write - 4 Threads : Ubuntu amd64, 2.6.30
                      SQLite : All the Ubuntu ones.
                      PostgreSQL: Solaris
                      It's not "missing" it's just that the bars there are too small proportionally to fit the text that shows what the exact value was. It's all charted correctly though.
                      Michael Larabel
                      http://www.michaellarabel.com/

                      Comment


                      • #12
                        Originally posted by parapup View Post
                        better behavior under insane load, unreasonable expectations of backward compatibility etc. - mostly the type of things which you don't want to deal with.
                        Those are very debatable things

                        Comment


                        • #13
                          Is it normal that Phoronix always uses a non-stable development version of their benchmark suite for each test? I mean every time i read a test on this site, not the actual stable version of the suite is used, it is always the development version.

                          Comment


                          • #14
                            Originally posted by glasen View Post
                            Is it normal that Phoronix always uses a non-stable development version of their benchmark suite for each test? I mean every time i read a test on this site, not the actual stable version of the suite is used, it is always the development version.
                            Yes, almost always. Whether you're running a development or stable version of the Phoronix Test Suite, as long as the test profile version is the same you should pretty much experience the same results. The pts-core really doesn't impact the actual testing process so much. And the development version is "stable" by most peoples definition, it's just a matter of going through changes at the present time. As the lead developer of the Phoronix Test Suite and also the one that facilitates 99% of the tests at Phoronix.com, it works out quite well and I wouldn't use a development version of PTS if I wasn't certain it could be counted on.
                            Michael Larabel
                            http://www.michaellarabel.com/

                            Comment


                            • #15
                              Originally posted by parapup View Post
                              So we are saying Mac OS X > Linux > Solaris here? If you ask me that sounds weird. I would prefer running Solaris in a hostile production environment and would venture into running Linux only after exercising caution but I would never want to run OSX even on moderately busy production server no matter what.

                              You know of course that VFAT can perform very very very well in file system benchmarks, right?

                              It's also true that efforts to protect and manage data efficiently on file systems have significant overhead....

                              Think about that and then it's no longer a mystery why OS X seems to be faster then Linux or Solaris.

                              ========

                              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...

                              ===============


                              Also it's worth noting that ZFS is generally pretty slow. It's not a fast FS even though it certainly is a very capable and very secure file system.

                              Everything has a trade off.

                              Also microbenchmarks are good at comparing Linux vs Linux.. or OS X vs OS X, etc etc.. but they are limited when comparing OS vs Different OS.

                              Comment

                              Working...
                              X