Cut read latency in an existing relational database design
You are shown the database design behind a real service: a relational store that has become slow to read. Propose how you would change it to bring read latency down. Start by asking about the access patterns and the slow queries, then weigh the options: indexes, query and schema changes (including denormalization), read replicas, caching in front of the database, partitioning, and precomputed views. For each, say what it costs in write speed, consistency and operations, and how you would roll it out and measure it.
No answer yet. Write the first one; the best answer gets endorsed.
