Is my general understanding that a typical database management systems bypass file system correct? I understand that they manage their own space on disk and they write actual data and index systems like B tree directly into disk blocks bypassing any intermediate help from file system.
This assumes that root would provide the database user permission to directly read and write from disk blocks. In Linux, this is still easier as disk can be treated as a file.
Any pointer to real case studies will be greatly appreciated.
Most rely on the underlying file system for WAL etc: basically they outsource it to the OS.
Some DBMS support (Oracle, MySQL) “raw” partitions, but it isn’t typical. Too much hassle (see this chat about Postgres) because you still need WAL etc on your raw partition.