Lecture8.pdf

💡Recall

🖋 Notes

DBMS Architecture

The Memory Hierarchy

primary storage

• cache, main memory
• very fast access to data
• volatile
• currently used data

secondary storage

• e.g. magnetic disks
• slower storage devices
• nonvolatile
• main database
• disks:
  • sequential, direct access

tertiary storage

• e.g. optical disks, tapes
• slowest storage devices
• nonvolatile
• tapes
  • only sequential access
  • suitable od for archives, backups
  • unsuitable for data that is frequently accessed

because

• disks and tapes - significantly cheaper than main memory
• large amounts of data that shouldn't be discarded when the system is restarted

⇒ the need for DBMSs that bring data from disks into main memory for processing

Magnetic Disks

• direct access

• extremely used in database applications

• DBMSs - applications don't need to know whether the data is on disk or in main memory

• disk block:

  • sequence of contiguous bytes
  • unit for data storage
  • unit for data transfer (reading data from disk / writing data to disk)
  • reading / writing a block - an input / output (I/O) operation

• tracks

  • concentric rings containing blocks, recorded on one or more platters

• sectors

  • arcs on tracks

• platters

  • single-sided, double-sided (data recorded on one / both surfaces)

• cylinder

  • set of all tracks with the same diameter

• disk heads

  • one per recorded surface
  • to read / write a block, a head must be on top of the block
  • all disk heads are moved as a unit
  • systems with one active head

• sector size

  • characteristic of the disk, cannot be modified

• block size

  • multiple of the sector size

• DBMSs operate on data when it is in memory

  • block - unit for data transfer between disk and main memory
  • time to access a desired location:
    • main memory - approximately the same for any location
    • disk - depends on where the data is stored
  • disk access time:
    • seek time + rotational delay + transfer time
    • seek time
      • time to move the disk head to the desired track (smaller platter size ⇒ decreased seek time)
    • rotational delay
      • time for the block to get under the head
    • transfer time
      • time to read / write the block, once the disk head is positioned over it

• time required for DB operations - dominated by the time taken to transfer blocks between disk and main memory

• goal

  • minimise access time
  • for this purpose, data should be carefully placed on disk

• records that are often used together should be close to each other:

  • same block
  • same track
  • same cylinder
  • adjacent cylinder

• accessing data in a sequential fashion reduces seek time and rotational delay

• characteristics:

  e.g.

  • storage capacity (e.g. GB)
  • platters
  • average / max seek time (ms)
  • average rotational delay (ms)
  • number of rotations / min
  • data transfer rate (MB/s)
  • ...

Moore's Law

"the improvement of integrated circuits is following an exponential curve that doubles every 18 months" by Gordon Moore

• parameters that follow Moore’s law
• parameters that do not follow Moore’s law

Solid-State Disks