Slideshow:
The need for different implementation of the
same algorithm
Overview of the
categories of data processing algorithms
|
One-pass algorithms
Two-pass algorithms
Multi-pass algorithms
Guideline for selecting algorithms
Also consider: index !!!
- Fact:
- There exists different implementation (categories) for the same query processing operator !!!!
- Reason
to
use different algorithms:
- Some algorithms use
a low number
of disk I/O operations,
but requires a
large amount of
memory space
- Other algorithms use a larger number of disk I/O operations, but requires a small amount of memory space
Therefore:
- In order to optimize the performance, we must choose the most suitable algorithm given the amount of resources in the computer system
- Some algorithms use
a low number
of disk I/O operations,
but requires a
large amount of
memory space
- The algorithms in
query processing can be broadly
catagorized as follows:
- One-pass algorithms:
- Read input relations
just once
- They use the lowest disk I/O
operations
- But, they
also
have the highest memory space
requirements
(They store one relation in memory !!!)
- Read input relations
just once
- 2-pass algorithms:
- Read input relations
2 times:
- The first time, the
operation performs a
preparation step
(e.g., sort)
and
write the
result to disk
- The second time, the prepared result is read and the actual operation is performed
- The first time, the
operation performs a
preparation step
(e.g., sort)
and
write the
result to disk
- They use a large number disk I/O
operations
- But, they
also
have a lower memory space
requirements
(They no not store entire relation in memory.
Only a portion of the relation is read)
- Read input relations
2 times:
- Multi-pass algorithms:
- Read input relations
more than 2 times
- They use the highest disk I/O
operations
- But, they also have the lowest memory space requirements
- Read input relations
more than 2 times
- One-pass algorithms:
- General guideline:
- If you have
sufficient memory space
to run a
one-pass algorithm, do that.
- Because one-pass algorithms are the most efficient algorithms !!!
- Otherwise,
check if
you have
sufficient memory space
to run a
2-pass algorithm.
If so, do that.
- Because the two-pass algorithms are the next most efficient algorithms
- Otherwise, run a
multi-pass algorithm.
- Multi-pass algorithms are the
least efficient
- But multi-pass algorithms can be used on input relations (files) of any size !!!
- Multi-pass algorithms are the
least efficient
- If you have
sufficient memory space
to run a
one-pass algorithm, do that.
- Note:
- These general guidelines may need to be modified if you can use an index to speed up the access to the tuples/records !!!