Hacker News Reader

Comment by fortran77 | Hacker News Reader

From the very start, the NT micro kernel was designed with multiple OS APIs in mind. They implemented a Posix subsystem decades ago.

View on HN

I think it's a little sad that they are going this route (wrapping a running Linux kernel) rather than working to improve Windows disk performance and continuing to improve their WSL 1.x product. It looks like they are missing out on an opportunity to improve the NT kernel for what looks like short-term gain.

From what I understand (which is very, very little!), fundamental NT vs Linux architecture differences wrt their respective file systems prevent much more performance improvement in WSL 1.x.

They do, but NT file performance over many files scales abysmally. The two major reasons are a legacy of bad decisions: deleting files is forbidden if there are open write handles, and filter drivers are allowed to be placed between other drivers in the file I/O stack.

There's not much Windows can do about this, which causes tons of issues with e.g. git clone. Erick Smith is ludicrously smart, he just couldn't squeeze better FS perf out of WSL1 due to these issues.

I wouldn't say bad decisions so much as different decisions. Given WSL would abstract out file system access anyway, why didn't they just bridge WSL's FS similarly to how they did with WSL2 anyway? Though, I'm not sure there's much advantage left at that point.

I don't think they could improve I/O speed without modifying NTFS itself (unlikely). This new approach is certainly easier to maintain and develop

You could implement ext4 on the nt kernel and have wsl1 run on it

Reply on news.ycombinator.com