Day 4 - ElastiCache and In-Memory Data Stores: The Physics of Speed

The difference between the time a database takes to read data from disk versus from memory ranges from thousands to tens of thousands of times. An NVMe SSD's sequential read latency is 50-100μs, while DRAM's access latency is 50-100ns. A 1000× difference. This physical gap is the reason caches exist. ElastiCache Redis, Memcached, DAX, MemoryDB — all of these services are built on top of this physical fact.

There are numbers that show in-memory data stores aren't merely a performance option. According to AWS, an architecture that leverages ElastiCache can reduce read latency by over 90% compared to an RDS-only setup and cut DB RCU cost by 60-80%