Clearly the best solution is for Microsoft to (a) write a proper ext4 driver for Windows and (b) find some way of embedding SIDs into ext4, then you could just format the drive as ext4, boot off it, and have the improved performance.
(This is mostly a joke, but the performance of NTFS for certain operations has always been abysmal, and having a virus scanner injecting itself into all the operations only makes it worse.)
AFAIK the main problem is that Unix's file permissions do not cover Windows' permission model. That would be tolerable on a data partition, but a system partition is going to use all kinds of very particular permission setups on system binaries etc.
You might be able to model that stuff as xattr, but then it could be problematic to mount that ext4 partition into Linux because applications might be copying files without respecting the xattrs.
>AFAIK the main problem is that Unix's file permissions do not cover Windows' permission model.
Well, since Microsoft has been borrowing more and more ideas from the Linux ecosystem, it would not surprise me that a Windows 10 successor would include some kind of compatibility layers for different file systems.
Why don't they just replace Windows with their own Linux distro? :D WSL2 cannibalizes Windows from the inside out, and all that's left is Sphere. Seems like the most efficient solution.
I wonder if this is like the IBM PC, which was a GOOD THING invented by a sort of an offshoot of IBM culture. Then IBM higher-ups stepped in and tried to control the platform (PS/2, OS/2, microchannel, etc)
WSL is attracting people to windows. But the endgame isn't to lose them to linux. So they have to tie it into windows more. But if they make it too slow and bloated they might lose.
Much of the performance problem comes from layers on top of NTFS itself- it's not just the virus scanner. Ext4 might be faster but I doubt it would be enough to ditch WSL2 for those use cases that need it.
Also, some of the "performance problems" are simply different access models. Windows and NTFS tries to provide some database-like ACID characteristics, including transactions at the level of batches of file updates with commit/rollback support. Ext4 and Linux (intentionally) make few such guarantees and so it shouldn't be surprising have very different performance profiles, just as you might expect between a NoSQL database that makes no ACID guarantees and an SQL database with multiple types of locks and several types of transaction behaviors.
(This is mostly a joke, but the performance of NTFS for certain operations has always been abysmal, and having a virus scanner injecting itself into all the operations only makes it worse.)