Slideshow:
Important observation about query processing
Important question: do we count the disk write operations needed for output ???
How to
pass data between
secessive operators:
materialization
Materialization:
In materialization: we count the disk write operations for output as cost
How to
pass data between
secessive operators:
pipelining
Pipelining:
In pipelining: we do not count the disk write operations for output as cost
Approach when computing the
cost of
an operation
- Fact:
- Processing a
query requires
a pipeline (= sequence) of
operations:
Important observation:
- The output of an operation is also the input of a subsequent operation
- Processing a
query requires
a pipeline (= sequence) of
operations:
- The are 2 different way to
pass the
output (result) of
an operation to
another
operation:
- Materialization:
- The output of the
operator is
written
(as a temporal relation) to
disk
- The receiving operator will
read the
tuples from disk
Note:
- This method will increase the number of disk IO operations
- The output of the
operator is
written
(as a temporal relation) to
disk
- Pipelining:
- The output of the
operator is
stored
in
memory buffer(s)
- The receiving operator will
read the
tuples from the memory buffer
Note:
- This method will not use any addition disk IO operation
- The output of the
operator is
stored
in
memory buffer(s)
- Materialization:
- Recall:
- Cost of an operator = the number of disk I/O operations (= # disk blocks) performed by the operator
-
Default assumption
used
in computing the
cost of an
operator:
- The output of the
operator is passed
using
memory buffers
- I.e.:
- The cost of an operation does not include the disk I/O's to write result (to disk)....
Note:
- When the cost computation
specifically
require that
an operation writes
its output to
disk, then:
- We will add in (= include) to disk IO cost for the output to the total cost
- The output of the
operator is passed
using
memory buffers