Denormalization involves what type of operations on normalized tables?

Denormalization primarily involves combining or splitting tables in order to optimize performance and reduce the complexity of data retrieval. When tables are normalized, the goal is to eliminate redundancy and ensure data integrity by structuring the data into smaller, related tables. However, this can lead to a situation where retrieving data requires multiple joins, which can adversely affect performance.

By denormalizing, you might combine several normalized tables into a single table that contains all the necessary attributes for specific queries. This reduces the need for joins and can lead to faster access times for read operations, which is a common goal in data warehousing and reporting scenarios. It is a strategic trade-off where developers may sacrifice some aspects of data integrity for improved performance.

The other options provided do not specifically capture the essence of denormalization. Creating new tables usually refers to expanding the database design rather than modifying existing structures. Removing duplicate data aligns more closely with normalization processes, which focus on ensuring that data is stored efficiently and non-redundantly. Indexing data also pertains to performance optimization but is a separate process that does not directly involve altering the table structure through denormalization.