Database Reference
In-Depth Information
• Sometimes the original file format in which data is stored does not provide
the required performance. The possible solution here is to create a new table
with a different file format or compression, and then use the
INSERT
state-
ment to perform a one-time conversion. This new table will provide compar-
atively better performance if you have chosen a new format or compression
carefully.
• Processing data, which is compressed, requires disk I/O and CPU cycles to
read and uncompress. However, if data were uncompressed, only the disk
I/O would comprise the primary cost during processing. So if the application
architecture supports processing, uncompressed data does expedite the per-
formance. With uncompressed data storage, you will end up taking lots of
space on the disk compared to compressed data. So, you will need to take
storage cost into consideration with performance gain.
• Sometimes, changing the file format or compression does not yield any per-
formance gain; rather it slows down the processing comparatively. In this
scenario, just using the original file and compression format is fine. So, the
lesson here is to understand the file and compression formats properly and
then choose them to derive better performance.
Tip
Chapter 7
,
Advanced Impala Concepts
, has more information about vari-
ous file formats and compression types and how to use them in Impala.
