# Google AlloyDB
Google AlloyDB is a fully managed, PostgreSQL-compatible database designed for
demanding enterprise workloads, specifically Hybrid Transactional and Analytical
Processing (HTAP).
![[arch_cluster_alloydb.png|256]]
## Intelligent Multi-Layer Caching
While AlloyDB shares the decoupled storage/compute philosophy of other
cloud-native databases, its crowning innovation is its **intelligent,
self-managing caching hierarchy**.
It uses active machine learning models to analyze query patterns in real-time,
deciding exactly which data should live in the fast compute memory ("hot") and
which should be relegated to the persistent storage layer ("cold").
## The Columnar Engine
Standard PostgreSQL stores data exclusively in a row-based format, which is
excellent for transactional inserts (OLTP) but terribly inefficient for scanning
large datasets for analytics (OLAP).
AlloyDB features an integrated, aggressively cached **Columnar Engine**. Its
machine learning models automatically identify data that is frequently accessed
for analytical reporting and fluidly converts those specific columns into a
highly optimized, in-memory columnar format. This allows AlloyDB to seamlessly
accelerate analytical queries by up to 100x without requiring a separate data
warehouse or manual ETL pipelines.