In the talk "Biggish Data With Rails and Postgresql" at RailsConf 2014, Starr Horne explores the challenges of managing large datasets in Rails applications using PostgreSQL. The presentation focuses on how traditional ActiveRecord conventions become ineffective as databases grow, leading to performance degradation. Horne shares his experiences from working at Honeybadger, a company that monitors errors and uptime, and emphasizes that common Rails practices can falter as data volume increases. The key points discussed include:

• Understanding Biggish Data: Horne clarifies that biggish data doesn't involve complex solutions like Hadoop but refers to practical issues faced as simple applications scale up to handle large data volumes.
• Performance Considerations: The speaker discusses how the performance characteristics of a database change significantly as its size increases. For example, typical pagination systems may fail at high page numbers due to inefficiencies in how queries are executed.
• Query Optimization: Horne emphasizes the importance of refining queries to limit the number of touched rows. He introduces the EXPLAIN command in PostgreSQL as a tool for understanding query performance and suggests using range queries to avoid the pitfalls of inefficient pagination.
• Infrastructure Improvements: The talk covers practical steps to enhance database performance, such as increasing RAM, optimizing disk I/O, and utilizing connection pooling to manage database connections efficiently.
• Database Management Techniques: Key strategies for maintaining performance over time include regular database vacuuming, partitioning data to improve deletion processes, and implementing read replicas for intensive queries.
• Backup Strategies for Large Datasets: Horne highlights challenges in backing up large databases and recommends using tools like WAL-E for incremental backups.

Ultimately, the speaker reassures the audience that while managing biggish data can be daunting, many issues arise one at a time and are solvable with practical approaches. The overarching message is that proper preparation and proactive measures can significantly ease the transition into handling large datasets, making applications more robust and performant.