Is my general knowning that an average database management systems bypass file system correct? I realize they manage their very own space on disk plus they write actual data and index systems like B tree straight into disk blocks skipping any intermediate the aid of file system.

This assumes that root provides the database user permission to directly read from disk blocks. In Linux, this really is still simpler as disk may be treatable like a file.

Any pointer to real situation studies is going to be greatly appreciated.

Most depend around the underlying file system for WAL etc: essentially they delegate it towards the OS.

Some DBMS support (Oracle, MySQL) "raw" partitions, however it is not typical. An excessive amount of hassle (see this chat about Postgres) since you still need WAL etc in your raw partition.

Not completely, mysql requests data directory and stores the information inside a specific file format where it attempts to optimize the reads,creates from the file as well as stores indexes there.

More over additionally, it may vary from one storage engine to a different. Mongodb uses memory planned files for disk IO Searching forward for additional discussion here.

DBMSs don't bypass filesystem. If the was the situation, the table names wouldn't be situation-insensitive under Home windows and situation-sensitive under Linux (in MySQL). The things they're doing would be to allocate large space around the file system (incidentally, the information continues to be visible like a file / group of files within the underlying operating-system) and manage internal data structure. This lower the fragmentation and also the overall overhead. Similarly cache systems works - Varnish allocates entire memory it requires having a single call to operating-system, then maitains internal data structure.