The Snowflake architecture is a hybrid of shared-disk (A common disk or storage device is shared by all computing nodes) and shared-nothing (Each computing node has a private memory and storage space) database architecture in order to combine the best of both. Snowflake utilizes a central data repository for persistent data, which is available to all compute nodes similar to a shared-disk architecture. But, equally, as with shared-nothing architectures, Snowflake uses massively parallel computing (MPP) clusters for query processing, in which each node stores part of the whole data set locally.
The Snowflake architecture is divided into three key layers as shown below:
Database Storage Layer: Once data has been loaded into Snowflake, this layer reorganizes that data into a specific format like columnar, compressed, and optimized format. The optimized data is stored in cloud storage.
Query Processing Layer: In the processing layer, queries are executed using virtual warehouses. Virtual warehouses are independent MPP (Massively Parallel Processing) compute clusters comprised of multiple compute nodes that Snowflake allocates from cloud providers. Due to the fact that virtual warehouses do not share their compute resources with each other, their performance is independent of each other.
Cloud Services Layer: It provides services to administer and manage a Snowflake data cloud, such as access control, authentication, metadata management, infrastructure management, query parsing, optimization, and many more.
