Presented by Russell Myers

advertisement
Presented by Russell Myers
Paper by Ming-Chuan Wu and Alejandro P. Buchmann
Problem
Optimizations and tuning designed for
On-Line Transaction Processing rather
than Data Warehouses
 Data Warehouses have specific needs

 Complex queries
 Huge volume of data returned as well as
updated

Database systems generally not built for
image data
Data Warehousing

How do we retrieve data efficiently
based on various metrics?
 Represented as a portion of the tuple
 Represented in metrics tables
 Represented as a defined data range (e.g.
bitmaps)
Tuple-based

Including the metrics directly in the tuple
 Hard to index based on and search through
 Not easily updated and costs a lot of storage
space (e.g. have to insert new metrics for
every tuple)
 Traverse in linear time
Metrics Tables

Data could be represented in
associative metrics tables
 Less data to store – just associations and
the metric data
 Quicker search – follow the metric to all the
primary keys of the individual pieces of data
 Still linear
Bitmapping

Each tuple contains a small portion of
encoded bitmap data
 Can organize the bitmap into decision trees
and use to search – less time
 Insertion and amount of data is lessened
 Slice the data into various portions to
represent multiple metrics and search
through those
Encoding Techniques

Hierarchical encoding
 Ex: Region to district to store

Range
 Eliminate searching from things outside of a
range (e.g. searching between 10 and 13
and excluding all others)

Ordering
The Paper
The paper seeks to prove that
bitmapping techniques are the way to go
 Performance analysis of various
bitmapping techniques
 Proofs of operations times

Critique

Excellent start
 Explained the topic clearly
 Made the basis of the paper easy to
understand

Confounding expansion
 Confused by direction
 Differences between various bitmap
techniques?
Critique
Bit slicing
 Mapping techniques
 Representations of bitmaps
 Metric graphical representation
 Advantages to different encoding
techniques

Contributions

Bitmap indexing and tree structures
bring faster data results
 Much cleaner than alternative means
 Easily represented
 Mapping situations are smaller than
representing each piece of data
Download