No, NTFS is cruddy. It is eager to flush metadata but not data, and there is no way to switch it to "journal data as well" mode (at least not as far as Win8.1,

No, NTFS is cruddy. It is eager to flush metadata but not data, and there is no way to switch it to "journal data as well" mode (at least not as far as Win8.1, maybe 10 added this mode). This is the default behavior on ext3 and ext4 as well, of course, and if you pull the plug on NTFS, ext3 or ext4, you WILL end up with files whose length reflects recent writes but whose contents do not (and I am saying this from analyzing hundreds of cases of file contents that are impossible given program flow -- about 10:1 on NTFS than on non-data-journalled ext4). But at least on linux you can tune2fs -o journal_data, still get very decent performance and much, much better integrity. Also, I think you are conflating sync(), fsync(), fdatasync() and FlushFileBuffers. calling fsync() or fdatasync() on linux is no worse than FlushFileBuffers - the IO killing behaviour you describe is associated with sync(), which, when you need it, is perfectly justified in killing IO.