Database Reference
In-Depth Information
Chapter 6. Denormalizing Data for
Maximum Performance
In the previous chapter, we created a structure that allows a user to follow other users. The
goal of the follow system was to allow users to see all of their followed users' status up-
dates in one place that we'll call the "home timeline". In this chapter, we will build a table
to store users' home timelines.
denormalization, the practice of storing the same piece of data in more than one place in or-
der to optimize read performance. The denormalization we used for follows was fairly
mild, however, each follow relationship is stored in exactly two places. For home timelines,
we will create a much more aggressively denormalized data structure: a given piece of data
will be stored in an arbitrary number of places.
While this highly denormalized structure will be the end result of our work in this chapter,
we'll explore several approaches along the way, starting with a fully normalized data struc-
ture, proceeding through a partially denormalized approach, and finally settling on a fully
denormalized design. We'll discuss the advantages and disadvantages of each, both in terms
of implementation complexity and runtime performance. By the end of this chapter, you'll
learn:
• How to retrieve data from more than one specific partition
• How to order and paginate in multipartition queries
• How to apply different denormalization strategies to a problem
• How to group multiple write statements into a single operation using logged
batches
