Announcement

Collapse
No announcement yet.

Windows 10 WSL vs. Docker on Windows 10 vs. Bare Metal Linux Performance

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

  • #11
    Originally posted by AndyChow View Post
    That the x264 encoding is faster on WSL than bare metal linux makes little sense. It was done completely on the CPU, right? I'm thinking maybe it's a turbo-power thing where Windows 10 just handles power states better for that particular chip than on linux. Intel Turbo boost technology doesn't sound like something linux would master right away.
    This was also true for C-Ray on Ubuntu. I wonder if WSL does some memory allocation tricks (i.e just grows the WSL memory pool and never decreases it until the process quits) and that x264 might be using a lot of malloc()+free() in the hot path. Because it should be just CPU-bound and mostly 1:1 in performance. I have no real knowledge of WSL but does it use exactly the same binaries that Ubuntu uses or does it use their own repository so that they might be compiled with a different version of GCC or something like that?

    Comment


    • #12
      Originally posted by edwaleni View Post
      is Docker exposing the CPU instructions properly?
      It's not Docker, but Windows Hyper-V that would be at fault here. On Windows 10, Docker runs under Hyper-V.

      Comment


      • #13
        Hi - PM for WSL here:

        Originally posted by SystemCrasher View Post
        So Windows sucks at IO & filesystems

        Actually, no! Despite claims to the contrary, NTFS is pretty powerful, flexible, and supports a very sophisticated and expressive permission system that powers and protects the data of ~80% of all computer users world-wide.

        Heck, we even use NTFS to store all your WSL files: WSL emulates Linux FS semantics and permissions, and maintains additional per-file/folder metadata. Alas, this overhead currently slows File IO down somewhat, but we're working hard to remedy this issue in future Windows 10 releases.

        Originally posted by SystemCrasher View Post
        and sucks at process creation

        Ermmm … no. For pretty much all non-disk-IO-intensive tests, WSL generally performs as well as *NIX on native tin, and is even a little quicker in some cases. Windows has improved A LOT in many ways compared to earlier versions of the OS.

        Originally posted by SystemCrasher View Post
        sends all keypresses to MS

        No, this is not true. Nowhere close. For those interested in what telemetry we DO receive, this may be of interest to you:
        http://www.zdnet.com/article/windows...elemetry-data/

        Originally posted by SystemCrasher View Post
        comes with weird license terms and costs something

        Most licenses are weird. They're written by lawyers or for/by people with particular objectives in mind, e.g.:
        https://www.networkworld.com/article...-kill-you.html

        Originally posted by SystemCrasher View Post
        Really good reason to use Linux on bare metal, any day
        For you, perhaps. But many can't run Linux natively due to relatively limited hardware/driver support, etc., or their employer has policies against doing so, or any number of other reasons. Others just prefer to run Windows. Thankfully, one can choose what to run, where and when.

        Comment


        • #14
          Originally posted by F.Ultra View Post

          I have no real knowledge of WSL but does it use exactly the same binaries that Ubuntu uses or does it use their own repository so that they might be compiled with a different version of GCC or something like that?
          Hi - PM for WSL here:

          WSL is distro agnostic. distro vendors like Canonical and SUSE upload their own distro to the store. The user downloads and installs the distros on their machines. WSL just executes unmodified ELF-64 binaries, translating Linux SYSCALLS into NT kernel calls where it can, or providing Linux specific behaviors/data where there's no native NT equivalent.

          For more info, take a look at this page full of resources: https://aka.ms/learnwsl.

          Comment


          • #15
            Originally posted by Michael View Post

            Each distro/OS file-system was their defaults.
            Hey Michael, it seems most of the performance impact WSL has comes from Windows Defender being enabled. I can confirm that disabling Windows Defender makes apt a lot faster on my computer. It seems Windows Defender will scan everything and slow everything down. If not too troublesome, could you try disabling Windows Defender the next time?

            Comment


            • #16
              Originally posted by andrebrait View Post

              Hey Michael, it seems most of the performance impact WSL has comes from Windows Defender being enabled. I can confirm that disabling Windows Defender makes apt a lot faster on my computer. It seems Windows Defender will scan everything and slow everything down. If not too troublesome, could you try disabling Windows Defender the next time?
              While it may produce faster results for WSL, please do not disable Defender or any 3rd party anti-malware in production environments. Doing so would be akin to driving without seatbelts. Better to get there a little later than not at all!

              The WSL team is working hard to remedy disk IO perf issues for future OS updates.

              Comment


              • #17
                Originally posted by bitcrazed View Post
                please do not disable Defender or any 3rd party anti-malware in production environments. Doing so would be akin to driving without seatbelts.
                What?

                Anti-virus and other anti-malware software does not have a demonstrable security benefit.
                In many cases, it makes your system even less secure (not surprising, running complex parsers on anything which the computer comes across...)

                https://int21.de/slides/33c3-evidencebased-infosec/

                While Microsoft Defender is often less problematic than 3rd party products, it is unnecessary to keep a system secure, and damaging to system security at worst.

                Comment


                • #18
                  Originally posted by chithanh View Post
                  While Microsoft Defender is often less problematic than 3rd party products, it is unnecessary to keep a system secure, and damaging to system security at worst.
                  My father in law - HOLD MY BEER!

                  While, yes, AV/AM may have reduced benefit to those of if who are conscientious about how we use our computers and what we allow to run on them, the VAST majority of unwitting users who -- yes, really DO want to watch this video and are happy for the necessary "Codec" to install to allow me to do so -- do not, and do not understand what they have to do to operate their computers safely.

                  For such users, an effective suite of anti-malware tools and regularly updated software and OS' are demonstrably beneficial. Not perfect - not by a loooong shot, but beneficial.

                  I've seen some stats demonstrating the efficacy of an anti-malware suite during a major vuln attack - it's
                  pretty compelling
                  and eye-opening stuff.

                  Comment


                  • #19
                    You have seen some stats that compare anti-malware vs. no anti-malware? Surely they can be incorporated into a peer reviewed scientific paper, but until now there are no such publications. Also the experts' opinion seems to not value anti-malware highly.

                    (Source: Google)

                    Originally posted by bitcrazed View Post
                    For such users, an effective suite of anti-malware tools and regularly updated software and OS' are demonstrably beneficial. Not perfect - not by a loooong shot, but beneficial.
                    Note that I, and the author of the slides which I linked to, use this term in a very narrow sense. Namely that the benefit has been demonstrated in a controlled randomized double blind trial.

                    Until then, the only sure "benefit" is that anti-malware serves as blameware, deflecting blame from users and management when an incident happens.

                    Comment

                    Working...
                    X