F-IVM: analytics over relational databases under updates

Area of Science:

• Database Management
• Data Analytics
• Machine Learning

Background:

• Maintaining analytics over evolving relational data is computationally challenging.
• Existing incremental view maintenance methods struggle with complex queries and large datasets.

Purpose of the Study:

• To present F-IVM, a novel unified approach for efficient analytics maintenance on changing relational data.
• To demonstrate the versatility of F-IVM across diverse computational tasks.

Main Methods:

• F-IVM employs higher-order incremental view maintenance, factorized computation, and ring abstraction.
• It unifies tasks by abstracting key-value structures and defining ring operations for payloads.
• Implementation on DBToaster facilitates efficient computation over keys, payloads, and updates.

Main Results:

• F-IVM successfully handles group-by aggregates, joins, linear regression, Chow-Liu trees, and matrix chain multiplication.
• The approach significantly outperforms classical incremental view maintenance methods.
• F-IVM demonstrates orders-of-magnitude performance gains and reduced memory consumption.

Conclusions:

• F-IVM provides a unified and highly efficient framework for maintaining analytics over changing relational data.
• Its factorized computation and ring abstraction offer a scalable solution for complex data analysis.
• The method shows substantial improvements over existing techniques, paving the way for more efficient data management.