Earlier quoted context omitted.
a local filesystem that can only be mounted once and of which state is 100% controlled by one entity (the local kernel) is not a distributed system the CAP theorem simply does not apply Windows IO subsystem was simply designed for extensibility over performance
A) It's a useful analogy whether or not you think it technically applies or is a perfect analogy. B) The Windows filesystem (and to an extent the POSIX) isn't just "local", it also includes transparent and semi-transparent network file storage. C) Windows and POSIX are both multi-user and multi-process . They operate over multiple cores and multiple I/O buses. Even if it just one system API centralized in charge of a…
it is almost the definition of "not a distributed system"
the fact there might be nfs/smb/ceph/... volumes bolted into the same namespace that happens to include it does not make it one (and neither does requiring transactions)
> Windows and NTFS combined have strong transaction guarantees that other users and processes must see a highly consistent view of the same files. It makes heavy uses of locks by default even for cached data. POSIX favors the "inode" approach that favors high availability and fewer locks at the cost of eventual consistency and the occasional partition (the "same" file can and will sometimes have multiple "inodes" between different processes/users, many common Linux tools rely heavily on that).
this is literally is not true, they are different abstractions
and I suggest you observe the size of a large file being copied if you want to see how "strong" NTFS "highly consistent views" are
https://devblogs.microsoft.com/oldnewthing/20111226-00/?p=88...