Ways to reduce the memory footprint of connections

Postgres has a high degree of redundancy in its catalogs and backends due to various constraints and design decisions. In previous talks I've gone into some detail on how we can improve memory usage in shared memory, but haven't yet gone into much detail about how we can improve session-level resource usage.

In this talk I'll go over some opportunities I've found that -once implemented- could reduce the memory footprint of connections by a healthy margin.