Ubuntu 14.04/16.04 vs. Ubuntu Bash On Windows 10 Anniversary Performance

[slacka]

What's the deal with the Dolfyn performance regression between 14.04 and 16.04? Is this a known issue? Has it been reported on the gcc bug tracker?

[Xaero_Vincent]

Is it possible to add the "Ubuntu 16.04 In VirtualBox On Windows 10" / "Ubuntu 16.04 In Hyper-V On Windows 10" compare to "Ubuntu 16.04 In VirtualBox On Windows 10 (Windows Subsystem for Linux)"

The comparation of memory usage and IO performance can be interesting

Windows Subsystem for Linux is much more like Wine than a VM and like Wine has flaws with app compatibility. However, I'd be interested in a VMware Player vs Virtualbox vs Xen vs KVM running a Windows 10 guest on a Ubuntu host versus native Windows 10 on bare metal.

[mike4]

How about running steam Linux games on Win10?

[smitty3268]

That x264 result has me fascinated. The only thing I can think of is that the SIMD support in Windows 10 has some Intel secret sauce that hasn't made it to the world of FOSS. I know Intel provided MSFT a lot of help with SIMD performance explicitly for Skylake, but with WSL overhead I wouldn't expect it to outperform native.

Isn't x264 sensitive to cpu scheduler algorithms? Could be that's the difference and all the threads it creates are getting a higher priority under windows or it's better at keeping them from interfering with each other, migrating across cores and losing caches, etc.

Also, the point of WSL is to run the apps natively so it's not surprising that there isn't much overhead on cpu bound tasks. The linux filesystem has to be emulated, so that's why it's so comparatively slow.

[Hi-Angel]

Overall, it shows that GNU/Linux got quite a room for optimizations. Running faster with API translation than natively — even a tiny bit — is a shame.

[Luke_Wolf]

Guys... This isn't virtualized anything, so other than the filesystem results... all of the results are as should be expected. Basically Windows like BSD has a pluggable ABI system of which Win32 is one of it's plugins, they implemented the Linux Kernel ABI and made it so that when dealing with that ABI the filesystem looks different to the running programs. It's possible that the redirection is what's causing the speed issues, or it could be that NTFS is just that bad in comparison.

[bug77]

Overall, it shows that GNU/Linux got quite a room for optimizations. Running faster with API translation than natively — even a tiny bit — is a shame.

It shows what you want it to show. More often than not, when something runs better with a translation layer than without, it means the translation layer is not doing everything it should (e.g. properly sync disk writes, support all video draw calls).
That said, yay for a proper CLI in Windows.

[Hi-Angel]

It shows what you want it to show. More often than not, when something runs better with a translation layer than without, it means the translation layer is not doing everything it should (e.g. properly sync disk writes, support all video draw calls).
That said, yay for a proper CLI in Windows.

Everything it should do is a proper support for API, and it does that, otherwise app wouldn't run. Improper sync disk writes would hurt themselves, so obviously they wouldn't do that.

[Xaero_Vincent]

How about running steam Linux games on Win10?

Considering that there isn't any 3D acceleration with this, I doubt that would be useful and the games would probably crash and burn before even rendering at 2 FPS.

However, why focus on bringing Linux apps into Windows when you can focus on bringing Windows 10 apps and interesting features straight into bare metal Linux?

I've shown this before but specialized virtualization paired with some handcrafted scripts can be a powerful combination and gives Microsoft a backseat position and Linux the front seat, where things ought to be:

[Nille]

but with WSL overhead I wouldn't expect it to outperform native.

There is no real overhead (exept for the Filesystem Stuff). WSL is only a Subsystem like Win32.