⚡Snowflake⚡ Basic Architecture Explained Simply: Storage, Compute, and Cloud Services
📚 Related Books
※ This article is reconstructed based on parts of a book. More in-depth content (such as design and specific examples) is compiled in the book, so please check it out if you are interested.
As a "first book" for those new to Snowflake.
As a "book to organize your thoughts" for those who use it somewhat vaguely but feel uneasy.
It is a book that will serve as solid fuel for becoming an engineer in the AI era.
Pattern 1: Understanding the "3-Layer Architecture" that supports Snowflake 🚀
When learning Snowflake for the first time, many people wonder, "Why is it so fast?" and "Why is it so easy to scale?" The answer lies in Snowflake's unique 3-layer architecture.
Snowflake is designed to scale independently by completely separating "storage," "compute," and "cloud services." In traditional data warehouses, storage and computing resources were closely linked, so both had to be increased together to improve performance. However, with Snowflake, you can flexibly increase or decrease only the resources you need.
Understanding this 3-layer structure reveals why Snowflake can achieve high performance, flexibility, and operability. In summary, the first step in learning Snowflake begins with understanding this basic architecture.
Pattern 2: Why does Snowflake scale? ❄️
Snowflake's greatest feature is its cloud-native design that completely separates compute and storage.
Users do not necessarily need to increase compute just because data volume has increased, and conversely, there is no need to change storage capacity even when executing large-scale analytical processing. Furthermore, since multiple Virtual Warehouses can be used simultaneously on the same data, there is a major benefit that different workloads such as ETL, BI, and data science do not affect each other.
Behind the scenes, the cloud services layer handles authentication, metadata management, optimization, and transaction control, allowing users to focus on analysis without being aware of the infrastructure. In summary, this decoupled architecture is what supports Snowflake's scalability.
Pattern 3: Understanding Snowflake's three layers with a diagrammatic image ✨
Snowflake is designed with a very simple structure.
Storage Layer
Compute Layer
Cloud Services Layer
The Storage Layer safely stores and compresses data, while in the Compute Layer, Virtual Warehouses handle SQL execution and ETL processing. The Cloud Services Layer oversees authentication, access control, metadata management, and query optimization.
Through this division of roles, users can start only the compute they need and stop it when it is no longer needed. As a result, it achieves high performance, low cost, and high concurrency.
To understand Snowflake deeply, it is important to first grasp the 3-layer structure. In summary, all functions are built upon this basic architecture.
⚡Snowflake⚡ Basic Architecture Explained Simply: Storage, Compute, and Cloud Services
The first step to understanding Snowflake is the "3-layer architecture"
When starting to learn about Snowflake, many people wonder, "How is it different from traditional data warehouses?"
The reason Snowflake achieves high performance and flexibility lies in its cloud-native 3-layer architecture.
Storage
Compute
Cloud Services
The complete separation of these three is Snowflake's greatest feature.
In this article, we will organize the roles of each and explain in an easy-to-understand way why Snowflake is so high-performing.
Storage Layer: Storing data safely and efficiently
The Storage Layer is, as the name suggests, the layer that stores data.
Data loaded by users is stored on cloud storage and is automatically compressed, encrypted, and micro-partitioned.
Users do not need to manage the storage destination in detail.
The Storage Layer performs the following roles:
Data storage
Compression
Encryption
Micro-partition management
Foundation for Time Travel and Fail-safe
Users can focus on "what data to handle" rather than "where to store it."
Compute Layer: Virtual Warehouse responsible for analytical processing
The Compute Layer is used when executing SQL in Snowflake.
Compute resources called Virtual Warehouses operate here.
Virtual Warehouses can be created freely for each purpose.
For example,
ETL Warehouses
BI Warehouses
Data Science Warehouses
can be separated and used.
What is even more important is that they all share the same Storage Layer.
In other words,
even if a large amount of ETL is executed,
BI users are not affected
allowing for high concurrency performance.
This is one of the reasons why Snowflake is strong in concurrency.
Cloud Services Layer: The brain that manages everything behind the scenes
The third is the Cloud Services Layer.
Although it is a part that is difficult for users to see, it plays an important role in controlling the entire Snowflake system.
Specifically,
authentication
authorization
Metadata management
Query optimization
Transaction management
Access control
SQL parser
and others operate here.
In other words,
it can be called the "command center"
that connects Storage and Compute.
Why does separating them improve performance?
In traditional data warehouses,
storage
compute
were integrated.
Therefore,
data volume increases
↓
compute also increases
was the structure.
However, with Snowflake,
you can manage storage capacity and compute performance independently.
In other words,
Only data increases
Only analysis increases
It can flexibly handle both.
This decoupled architecture is a design highly suited for the cloud era.
Virtual Warehouses support concurrent execution
In Snowflake, you can create multiple Virtual Warehouses.
For example,
Development team
BI team
Data engineers
Machine learning team
Each can use a separate Warehouse.
Even when looking at the same data,
the compute is completely independent.
Therefore,
heavy ETL
BI dashboards
ad-hoc analysis
are designed so that they are unlikely to affect each other even when executed simultaneously.
This is one of the reasons why Snowflake is adopted by many companies.
Simple operations because it is cloud-native
With Snowflake, you rarely need to worry about infrastructure management.
Users only need to:
choose a Warehouse size
start it only when needed
.
Snowflake handles cluster construction, OS management, and storage management.
As a result, you can achieve:
low operational overhead
automatic scaling
high availability
.
Understanding the 3-layer architecture gives you a clear view of Snowflake
Snowflake has many features such as:
Time Travel
Zero Copy Clone
Data Sharing
Snowpark
Dynamic Tables
.
However, everything is built upon this three-layer architecture.
In other words,
Storage, Compute, and Cloud Services
Once you understand these roles, it becomes easier to understand new features as well.
Summary
The most significant feature of Snowflake is its
Storage
Compute
Cloud Services
architecture, which completely separates these components.
This structure enables:
High scalability
High concurrency performance
Simple operations
Flexible cost optimization
are achieved.
When learning Snowflake, understanding this three-layer architecture first will allow you to smoothly grasp the many features and design philosophies that follow. In summary, the strength of Snowflake lies not in individual features, but in this simple yet powerful basic architecture.
📚 Related Books
*This article is reconstructed based on parts of a book. More in-depth content (such as design and concrete examples) is compiled in the book, so please check it out if you are interested.
I have compiled the "Learned by Touching from Scratch!" series, where you can systematically learn about Databricks, Snowflake, n8n, Salesforce, AI infrastructure, POC, and how to proceed with requirements definition.
"Learned by Touching from Scratch! Databricks Production Implementation Complete Guide (Unofficial) ― Data Infrastructure Practice in the Era of Serverless, Lakeflow, and AI ―"
"Databricks: Learned by Touching from Scratch! Unofficial Databricks Guide (2026 Updated Edition)"
A bestseller introductory book that allows you to learn cloud-era analytics infrastructure "experientially."
Covers Databricks operations, SQL/DataFrame, basics of Delta Lake, notebook operations, and SDP (Declarative Pipelines).
Explains Serverless, Genie, and more in a structure that even beginners can follow without getting lost.
https://amzn.to/4uIqEj4
"Learned by Touching from Scratch! Snowflake × Databricks Next-Generation Data Infrastructure PoC Practice Unofficial Guide"
By the time you finish this book, we aim for you to have a clear understanding of "what to start with for a POC" and the confidence that "we can do this too."
https://amzn.to/43qI0oR
"Learned by Touching from Scratch! Next-Generation Data Infrastructure Built with Snowflake × Databricks - Comparison, Coexistence, and Integration Unofficial Guide"
Snowflake and Databricks—these two cloud data platforms have often been discussed in terms of "which one to choose." Based on the actual experience of building and operating both platforms from scratch, this book is an "unofficial guide" that carefully explains the reality of their comparison, coexistence, and integration.
https://amzn.to/4efDkIk
Snowflake
Learned by Touching from Scratch! Unofficial Snowflake Guide ― Understanding Architecture from the Basics and Next-Generation AI Infrastructure with Cortex
As a "first book" for those touching Snowflake for the first time.
As a "book to organize your thoughts" for those who are using it vaguely but feel uneasy.
A book that serves as solid fuel for engineers aiming to thrive in the AI era.
https://amzn.to/4x1VvZm
"Learned by Touching from Scratch! Codex - Software Design in the Era of AI Agents"
This book provides guidelines for engineers to continue demonstrating value proactively without stopping their thinking in an era of developing alongside AI agents.
For all engineers who want to understand the essence of future development, not just how to use tools.
https://amzn.to/4o0repH
"Learned by Touching from Scratch! Claude Code × ChatGPT × Gemini AI Symbiosis Strategy - Toward an Era of 'Symbiosis' Rather Than 'Conflict'"
Explains the symbiosis model of Claude Code × ChatGPT × Gemini.
https://amzn.to/4a2dJjC
"Learned by Touching from Scratch! Introduction to Spec-Driven Development ― Is SaaS Dead? Software Design Theory in the AI Era"
The first half carefully organizes the philosophy and background, while the second half concretizes the three-layer model of spec, implementation, and execution with sample code.
https://amzn.to/3RFEZya
#### "Learned by Touching from Scratch! dbt Practical Introduction Unofficial Guide: Modern Data Transformation, Testing, Documentation, and Semantic Layer Built with SQL"
This book is an introductory guide written for those who want to learn dbt, those who want to manage data transformation using SQL more systematically, and those interested in data infrastructure and the modern data stack.
https://amzn.to/3SmxlJz
Databricks
"Learned by Touching from Scratch! Next-Generation Data Infrastructure Built with Azure × Databricks Unofficial Guide ―"
When you try to build a data platform in the cloud, you face Azure, Storage, networking, permissions, security...
The moment Databricks is added to that mix, the difficulty level jumps significantly. This book carefully unravels those "initial stumbling blocks."
https://amzn.to/3QaOzbW
"Databricks: An Unofficial Guide for AI and Machine Learning Engineers—Learned by Trying It from Scratch!"
The definitive introductory guide for AI engineers covering prompt design, RAG construction, model management, and governance in Databricks.
It systematizes the "practical patterns" necessary to bridge the gap between generative AI and data engineering.
This is not a certification book, but content designed to cultivate the ability to operate AI as a practical foundation.
https://amzn.to/3PYK4ku
"Databricks Certified Data Engineer Professional Exam Level — 3 Minutes a Day! Databricks You Can Read from Wherever You're Interested! 100 Soul-Stirring Knocks!"
This book breaks down topics equivalent to the Databricks Certified Data Engineer Professional exam into 100 use cases.
It is a "100 Soul-Stirring Knocks" series that rhythmically drills concepts into you through "2-choice considerations" → "explanatory columns" → "conclusions."
Rather than rote memorization, it is structured so you can repeat the decision points you encounter in the field (ingestion, transformation, quality, sharing, monitoring, performance/cost, security, governance, deployment, modeling) in short reading sessions.
https://amzn.to/4vkLm8K
https://amzn.to/4fhNBF5
Databricks Advanced Series (Vol. 1 / 2 / 3)
The complete practical guide for "designing and operating" Databricks.
Born as a sequel to "Databricks: An Unofficial Guide for AI and Machine Learning Engineers—Learned by Trying It from Scratch!," the Advanced Series is structured to step beyond just getting used to Databricks—diving into full-scale operation, team development, certification preparation, and reproducible design.
📘 [Vol. 1] Development, Deployment, and Quality Assurance
https://amzn.to/4dGQoGv📘 [Vol. 2] Ingestion, Transformation, Monitoring, and Cost Optimization
https://amzn.to/49zbPHb📘 [Vol. 3] Security, Governance, Troubleshooting, and Optimization Strategy
https://amzn.to/4efDkIk
"Databricks: An Unofficial Guide—Learned by Trying It from Scratch! Databricks × Airbyte"
A guide has been completed that carefully unravels why data platforms in the cloud era are "difficult."
It gently systematizes Ingestion, LakeFlow, DLT, and CDC, and is a book packed with the design philosophy to bring out the true value of Airbyte × Databricks.
https://amzn.to/3XOlV0t
"Databricks: An Unofficial Guide—Learned by Trying It from Scratch! Databricks and Confluent (Kafka) Integration!"
A hands-on book that integrates stream processing via Kafka with Databricks to build a real-time analytics platform.
It summarizes essential skills for modern data platforms, including event-driven architecture, real-time ETL, and Delta Live Tables integration.
https://amzn.to/42HdmqZ
Salesforce
"An Unofficial Guide—Learned by Trying It from Scratch! Salesforce AgentForce + Data360 (Data Cloud)"
An explanatory book where you can understand Salesforce's latest AI platform, AgentForce, and Data360 (Data Cloud) through actual operations.
https://amzn.to/4u4PyZ2
Requirements Definition (Upstream Processes / Modern Data Stack)
"The New Requirements Definition for the Modern Data Stack Era: A Complete Guide to Cloud Construction — From DWH to CDP, and then MA / AI Integration"
How should you think about "requirements definition" in the cloud era?
It gently explains the "overall picture of construction" while traversing major services such as Databricks, Snowflake, Salesforce, and n8n!
From DWH to CDP, and then MA/AI integration—get the practical knowledge you can use in the field in this one book.
https://amzn.to/4nZm0ux
Data Mesh
#### "An Unofficial Guide—Learned by Trying It from Scratch! Introduction to Data Mesh: Thinking about Data Mesh from Philosophy, Patterns, and Organizational Structure"
This is not a book that asserts whether or not you "should" introduce Data Mesh.
For yourselves, how far should you decentralize, what should you share, and where should you place responsibility?
It is a book that organizes the foundation of thinking to make those judgments.
https://amzn.to/3REkyBS
Data Clean Room
A Practical Introduction to Data Clean Rooms: Learned by Doing from Scratch — Decoding Clean Rooms in the Lakehouse Era through Philosophy, Design, and Monetization
You have the data, but you can't share it. Yet, you still want to analyze it together—this book begins with that common workplace dilemma.
We have organized the data clean room not as a 'difficult technology,' but from the perspective of how to use it and sustain it in real-world operations.
A practical book that even non-IT business professionals can read.
https://amzn.to/4fiG6O2
MCP
'MCP Beginners Guide: Learned by Doing from Scratch' — An Introduction to the Next-Generation Protocol for the AI Agent Era: Architecture, Governance, and Implementation
The MCP protocol is not just a technology trend;
it holds the potential to change the very 'relationship between AI and systems.'
This is a book that pauses to organize the intersection of SaaS, AI agents, governance, and architecture.
https://amzn.to/4nZm0ux
n8n
'n8n—Learned by Doing from Scratch! AI Workflow Automation! Unofficial Guide'
A practical guide to learning the open-source automation tool n8n by 'getting your hands dirty from scratch.'
Designed with a step-by-step, screen-operation-focused structure so that even those who are not good at programming can follow along,
you will firmly grasp the basics of business automation, AI integration, and API integration.
👉 https://amzn.to/48Blxca
💡 Summary: Gain the 'Builder's Perspective' with this lineup
Through these books,
you can systematically and practically acquire the big picture of a solution architect in the modern data stack era,
covering: Cloud Infrastructure Understanding → Requirements Definition → Analytics Infrastructure Construction → Automation → AI Integration → Operational Optimization.
PoC Requirements Gathering
Data Infrastructure Requirements Definition
Team Development / Governance
AI Workflow Construction
Troubleshooting
These can also be used as knowledge to solve challenges often faced in the field.