3 ms·
I'm going to defer to you on this one, but my hunch is that NTFS being stupid and crusty is a limiting factor? In my experience, it's dog-slow (particularly on
by mises 7y ago
I'm going to defer to you on this one, but my hunch is that NTFS being stupid and crusty is a limiting factor? In my experience, it's dog-slow (particularly on small files). Have you tested on ReFS and compared performance?
- apankrat 7y agoNo, I have not. From what I've seen NTFS is a really well-made file system. There is indeed a fixed per-file overhead that becomes an issue when working through lots of small files, but NTFS handles multi-threaded access very efficiently, so that's a very easy workaround.
- mehrdadn 7y agoIt's not NTFS that's slow, it's something in the Windows I/O subsystem common across file systems. And it's not that it's slow on small files (the I/O performance is fine), but that there's a large overhead for repeatedly opening and closing files, especially by paths. Apparently Linux gets around this with some sort of inode cache (hashtable?) whereas Windows doesn't implement this for whatever reasons. (Possibly the filter drive model used to support virus scanning etc. doesn't allow it, though I don't quite understand the details of why.) Not even sure Mac is as fast as Linux in this regard.
- rodgerd 7y agoOn the other hand we have, over the years, seen plenty of fights between application and Linux kernel devs over the degree to which the POSIX spec makes it OK for Linux to discard application data, so perhaps that speed comes at a cost.