---

Intro to Pointer Swizzling

---

---

---

 

Slideshow:

Inefficiency in the Translation table technique

The most efficient way to access records that are stored in memory

The most efficient way to access records that are stored in memory

Pointer "swizzling"

Pointer "swizzling" technique illustrated

Pointer "swizzling" technique illustrated

I.e.: the database address is replaced by a memory address !!
Implementing the pointer "swizzling" technique
Data Structure:

Implementing the pointer "swizzling" technique
Initial state:

Implementing the pointer "swizzling" technique
Suppose we read in Block 1 into the memory:

Implementing the pointer "swizzling" technique
Look up the database addresses in the translation table:

Implementing the pointer "swizzling" technique
Replace the database address with the (virtual) memory address:
                      
When do we perform pointer swizzling ?
This is a policy decision:

❮ ❯



---

---

---

---

• Inefficiency in Translation table
  • Inefficiency in accessing a block/record that is stored in memory using the translation table:

    We must translate the database address each time we want to access a block stored in memory

    ---

    ---

  • The most efficient way to access a record that is stored in memory is direct access using a memory address:

    Observation:

    To achieve better performance: We must replace the database address in a record by the virtual memory address of the record !!!

  ---

  ---

  ---

  ---

• Pointer swizzling
  • Pointer swizzling:

    Pointer Swizzling = the database addresses in data blocks/records are replaced by the corresponding virtual memory addresses when the referenced data block/record resides in memory

    Pointer swizzling illustrated:

    Suppose a record (in block 1) references another record in data block block 2 (stored on disk) Situation before we read the data block block 2 (stored on disk) into memory: The record stored in block 2 (on disk) is referenced by its database (record) address (because the block is only stored on disk) After the data block 2 is read into memory: The record stored in block 2 is referenced by its (virtual) memory address !!! This is the result of "pointer swizzling" It is most efficient way to reference data stored in the disk block block 2

  ---

  ---

  ---

  ---

• How to implement the pointer swizzle technique
  • The record field that contains a database address has the following structure:

    A database address field: +-+--------------------------------------------+ | | address | +-+--------------------------------------------+ ^ | Flag bit: 0 ==> field contains a database address 1 ==> field contains a virtual memory address

    (It's similar to the union structure of the C programming language)

    ---

    ---

  • When you first read a data block into the memory:

    The database address stored in the data block on disk are always database address:

    Procedure to swizzle a database address into a (virtual) memory address:

    Look up the database address in the translation table If found: Replace the database address in block/record with the (virtual) memory address in Translation Table Graphically:

  ---

  ---

  ---

  ---

• Swizzling Policies
  • Policies governing when to perform pointer swizzling:

    Automatic swizzling: When a block is read into main memory: we replace all database addresses in the records stored in the block On-demand swizzling: When a block is read into main memory: we do not swizzle (replace) any database addresses in the block When we access a record that contains a database address: If database address has not yet been swizzled: Swizzle the database address into a virtual address (using the translation table) Then access the database object