The article explores how the design of relational databases needs to evolve in light of advancements in SSD technology, cloud infrastructure, and modern application demands. With local NVMe SSDs offering a dramatic 1000x improvement in throughput and latency compared to traditional spinning disks, existing database architectures, like Postgres and MySQL, are no longer optimal. The author, Marc Brooker, emphasizes that a fresh approach is necessary to capitalize on these improvements. Brooker presents several key strategies for redesigning databases. The first revolves around the "Five Minute Rule," which suggests that RAM caches should be sized to retain pages expected to be accessed within 30 seconds, based on current EC2 instance costs. Next, he discusses the importance of transfer sizes; modern SSDs are efficient with transfers around 32kB. Transfers smaller than this can hinder throughput and increase latency. The third approach addresses durability—modern databases must replicate data off-box to ensure reliability, particularly across availability zones, while minimizing latency during commits. Finally, Brooker critiques traditional write-ahead logs (WALs) as unnecessary in a replicated database environment. Instead, he argues for designs that leverage multi-versioning and modern hardware capabilities, enabling databases to maintain strong consistency without the drawbacks of slower commit-to-disk processes. The discussion highlights the need for database vendors to rethink their architectures to better serve the high-performance requirements of contemporary applications.