Announcement

Collapse
No announcement yet.

EXT4 File-System Looks To Do Well Against NTFS

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

  • #16
    One thing that is important to note, are the drives running in AHCI mode or IDE Mode and if running in AHCI mode are you using the built in microsoft drivers or Intels chipset drivers. There is quite a performance difference between the modes and drivers.

    Comment


    • #17
      Originally posted by dashcloud View Post
      More interesting would be benchmarking NTFS-3g vs Windows 7 NTFS. I've heard the performance is very good with NTFS-3g, and there's apparently a souped-up embedded version that's crazy fast.
      That would be an interesting test.
      NTFS-3g seems to use an insanely poor block allocation strategy. I copied a couple hundred gigabytes of files, ranging in size from a few KB to several GB, to a fresly-formatted NTFS partition using NTFS-3g. Accessing the files was incredibly slow, so I ran the Windows defragmenter. 70% of the files had 2 or more fragments, and it was actually so bad that 20 files failed to defragment (a couple of them happened to be sparse VM images). All of the blocks were scattered around the center of the partition, negating the read performance benefits and lack of fragmentation one would expect to achieve by copying files onto a fresh partition.

      Comment


      • #18
        Originally posted by gilboa View Post
        As I recall, ext3 (out-of-the-box) was somewhat faster than a optimized ext4 (write barrier disabled, we were using a huge FC storage).
        And both were ~12-20x faster then 2K8 NTFS.
        - Gilboa
        That's been my limited experience - NTFS isn't just slow it's abysmal.

        That ext4 is 'on par' with it really points to ext4 being a very very poor choice. (and probably that the benchmark wasn't done very well).

        Comment


        • #19
          The ntfs-3g port to OpenSolaris is also slow. (Yes, "famous troll is back")

          Comment


          • #20
            Originally posted by kebabbert View Post
            The ntfs-3g port to OpenSolaris is also slow. (Yes, "famous troll is back")
            If ntfs is slow it's port will be slow too right? Ntfs-3g is probably noticeably slower then native. ("He didn't show papers ever, but he was considering all the time he did, it was so funny :>").

            Comment


            • #21
              +1 for the ntfs-3g idea. You could also benchmark the in-kernel ntfs performance for read-only tests. And maybe WinXP too?

              Comment


              • #22
                Originally posted by Flyser View Post
                And maybe WinXP too?
                WinXP? Lol!
                That's just pointless, if you really want to know the performance of NTFS test Windows Server.

                Comment


                • #23
                  Originally posted by Enrox View Post
                  When you run filesystem tests, please specify if Windows Defender or any other antivirus software is running, it makes a huge difference.
                  As a huge difference makes the nobarrier option of ext4 but then we go to other things etc etc. So every system stays in its default settings.
                  Actually, 26.5% difference is rather surprising. In reality I find ext4 almost twice as fast as ntfs.

                  Comment


                  • #24
                    Originally posted by Apopas View Post
                    As a huge difference makes the nobarrier option of ext4 but then we go to other things etc etc. So every system stays in its default settings.
                    Question is - who's Phoronix' "ideal" client?
                    As a large portion of the application that are being used during benchmarks, the claim that the sys-admin is stupid and could not (or should not?) be expected to flip a switch on certain performance improving switches in its distribution (Be that write barrier in ext4 or disable 8.3 in NTFS) is ridicules at best.

                    The ext4 developers cannot (and must not!) assume that you have battery backup on your HBA or good UPS, only the sysadmin can make that call and decide if the performance improvement is worth the additional risk.

                    BTW, take the time to compare ext3 in ordered mode (default) and journal mode. You'll be amazed by the performance hit.

                    - Gilboa
                    DEV: Intel S2600C0, 2xE52658V2, 32GB, 4x2TB + 2x3TB, GTX780, F21/x86_64, Dell U2711.
                    SRV: Intel S5520SC, 2xX5680, 36GB, 4x2TB, GTX550, F21/x86_64, Dell U2412..
                    BACK: Tyan Tempest i5400XT, 2xE5335, 8GB, 3x1.5TB, 9800GTX, F21/x86-64.
                    LAP: ASUS N56VJ, i7-3630QM, 16GB, 1TB, 635M, F21/x86_64.

                    Comment


                    • #25
                      Originally posted by gilboa View Post
                      BTW, take the time to compare ext3 in ordered mode (default) and journal mode. You'll be amazed by the performance hit.
                      I have
                      In the ext3 days though, I was using reiserfs.

                      Comment


                      • #26
                        Originally posted by Apopas View Post
                        As a huge difference makes the nobarrier option of ext4 but then we go to other things etc etc. So every system stays in its default settings.
                        It probably depends on tests. I ran some benchmarks using PTS and the difference was sadly quite small. I bet Ext3 will be much, much better in some PTS tests.

                        Comment


                        • #27
                          Originally posted by Apopas View Post
                          As a huge difference makes the nobarrier option of ext4 but then we go to other things etc etc.
                          Correct. Any important information regarding the configuration of the tested system has to be clearly known.
                          It might be ok to leave the default parameters of the file system itself, it make less sense not to turn off Windows Defender since it's not a "core component" of the file system.

                          So every system stays in its default settings.
                          Why? I don't think there's a martial law for that!
                          When you test performance you do it for a reason... to see what are the best numbers you can get out of a system and to know what parameters make a difference.
                          Basically no one in the world runs a system without making any change.
                          So what's the point of testing the default settings?

                          If someone is concerned about a system performance it means that is already tweaking the system.

                          Also... upcoming tests will use Apache, SQLite, PostgreSQL on Windows 7, those are not the dafault software used on Windows 7. No one is using Windows 7 to reach best numbers out of those softwares. It would make much more sense to test on Windows Server and not on the client.

                          Unless we assume that those benchmark are just pointless waste of time

                          Comment


                          • #28
                            I don't know much about file systems, but I see that many more technical users have reported that NTFS is much slower than EXT4, so wouldn't it be useful to include real world file system tests? Even basic ones like file creation, copy, delete, etc. This could probably enlighten less techical users (as myself) to the real advantages of one or the other in everyday usage.
                            Even in my ignorance on these matters I believe that something most be really well done on the EXT4 side since I installed ubuntu 10.04 on my eee pc the other day and it only took 10 minutes, where a normal windows xp instalation could take more then half an hour.

                            Comment


                            • #29
                              Originally posted by Enrox View Post
                              Correct. Any important information regarding the configuration of the tested system has to be clearly known.
                              It might be ok to leave the default parameters of the file system itself, it make less sense not to turn off Windows Defender since it's not a "core component" of the file system.
                              These tests have to do with the out of the box experience. We want to see the experience of the simple user who buys his computer with windows7 or ubuntu preinstalled since this kind of user rarely tweak things.
                              It's funny though that you speak for things that may hammer the windows performance, while Ubuntu can be tweaked in a lot of ways rather than just by disabling some processes...


                              Why? I don't think there's a martial law for that!
                              For sure not, but still Linux with it's glorious configurable ability can easily produce serious arguments about th options that were used instead of the others or the more others that were not, or even the distro, the DE etc etc.


                              When you test performance you do it for a reason... to see what are the best numbers you can get out of a system and to know what parameters make a difference.
                              These tests don't show the performance in general, but the performance in its default settings.
                              If pure performance was the goal then the choice wouldn't be Ubuntu but Gentoo with it's vast tweak abilities.

                              Basically no one in the world runs a system without making any change.
                              So what's the point of testing the default settings?
                              The vast majority of people never go thus far as to change something more than their desktop wallpaper, how much to change things that affect the performance of a filesystem or a web server... Don't think that the average computer user use to lie in forums like this one...
                              Hey that's what Apple promotes all these years and goes quite well. "Great experience out of the box".

                              If someone is concerned about a system performance it means that is already tweaking the system.
                              Not really. A lot of people suffer from low performance of their computers but have absolutely no idea what causes it and how can be fixed.

                              Also... upcoming tests will use Apache, SQLite, PostgreSQL on Windows 7, those are not the dafault software used on Windows 7. No one is using Windows 7 to reach best numbers out of those softwares. It would make much more sense to test on Windows Server and not on the client.
                              The tests show the performance differences between two desktop OSes.
                              It's normal to test Ubuntu with windows7. If windows server was the candidate, then normally it would be challenged by Ubuntu Server ot RHEL or Centos.

                              Unless we assume that those benchmark are just pointless waste of time
                              For a company or an individual who cares for the out of the box experience, these tests are rather valuable. For the user (like me) who likes to get the maximum performance, can take this specific results as the a metre of comparison.

                              Comment


                              • #30
                                Originally posted by Apopas View Post
                                For a company or an individual who cares for the out of the box experience, these tests are rather valuable. For the user (like me) who likes to get the maximum performance, can take this specific results as the a metre of comparison.
                                Here's where your argument completely fails.
                                Out of the box performance may be relevant when you're talking about desktop applications (to some extent) such as games and general utilities (lzma/gzip, mp3 encoding, etc).
                                However, out of the box experience loses all meaning once you start talking about server applications such as apache or postgresql and specialized applications such ray-tracing or simulation software.

                                - Gilboa
                                DEV: Intel S2600C0, 2xE52658V2, 32GB, 4x2TB + 2x3TB, GTX780, F21/x86_64, Dell U2711.
                                SRV: Intel S5520SC, 2xX5680, 36GB, 4x2TB, GTX550, F21/x86_64, Dell U2412..
                                BACK: Tyan Tempest i5400XT, 2xE5335, 8GB, 3x1.5TB, 9800GTX, F21/x86-64.
                                LAP: ASUS N56VJ, i7-3630QM, 16GB, 1TB, 635M, F21/x86_64.

                                Comment

                                Working...
                                X