Disk-oriented DBMS

• DB are stored as normal files on disk. OS doesn’t know about in file content.
  • Most DBMS runs on top of other file systems (like ext4, etc.)
  • Some DBMS stores single file (e.g. sqlite, duck), but most DBMS store as multiple files.
• Within the DB file there can be multiple pages
  • each page start with a header
  • there is a special structure (called directory) to store metadata about the pages
• in memory, there is buffer pool manager that controls which page to cache in memory
  • ensures that the page will be available in memory until execution engine done using it
• need 2 kinds of control:
  • spatial control: where to write pages on disk
  • temporal control: when to pull and flush pages to memory

DB representation in files

components

…

• Storage manager: maintaining DB files
  • some even schedule reads and writes to take advantage of locality
  • organizes files as a collection of pages
    • track reads and writes and
    • available space, etc.
  • doesn’t handle replication, etc. → ensure only unique copy of a page is stored
• DB pages:
  • A page is a fixed-size block of data with unique ID (within any scope, DB, table, DBMS, etc.)
  • This is different from hardware page
    • a hardware page is the largest block of data that storage device can guarantee failsafe!
  • read-heavy DB should have larger page (1MB+) to fetch multiple content at once
  • write-heavy DB should have smaller page (around 4KB) because DB needs to flush to disk even when only little portion is changed in the page
• DB file organization:
  • different DBMS stores pages in files in different ways:
    • heap file organization
    • tree file organization
    • sequential/sorted file (ISAM)
    • hashing file
  • Heap file: unordered collection of pages with tuples stored in arbitrary order
    • need additional metadata to track location of pages in files
      • DBMS maintains special pages called page directory
        • needs one per database
      • also additional metadata like remaining space, page type (data/metadata), etc.

DB page format

2 parts: header and data

Header

metadata about data part:

• page size
• checksum
• DBMS version
• compression scheme, …