Domain 1.1 — Snowflake Architecture
Master Snowflake's three-layer architecture: Cloud Services, Compute (Virtual Warehouses), and Database Storage. Understand how separation of storage and compute drives Snowflake's core value.
Domain 1.1 — Snowflake Architecture
Exam Weight
Domain 1.0 — Snowflake AI Data Cloud Features & Architecture accounts for ~31% of the SnowPro Core COF-C03 exam — the largest single domain.
This lesson maps to Exam Objective 1.1: Describe and use the Snowflake architecture, including the Cloud Services layer, Compute layer, Database Storage layer, and comparison of Snowflake editions.
What Is Snowflake?
Snowflake is a cloud-native, fully managed data platform delivered as Software-as-a-Service (SaaS). It was built from the ground up for the cloud — not ported from an on-premises product.
Key distinctions from traditional warehouses:
| Characteristic | Traditional Warehouse | Snowflake |
|---|---|---|
| Deployment | On-premises / IaaS | Pure SaaS |
| Scaling | Requires downtime | Seconds, zero downtime |
| Compute & Storage | Tightly coupled | Fully separated |
| Maintenance | Customer-managed | Snowflake-managed |
| Cloud providers | Single | AWS, Azure, GCP |
| Updates | Planned downtime | Automatic, transparent |
The Three-Layer Architecture
Snowflake's architecture consists of three independently scalable layers. Understanding each layer — and the boundaries between them — is critical for the exam.
Layer 1 — Database Storage Layer
This is where all data permanently lives. Key characteristics:
- Data is stored in Snowflake's proprietary columnar micro-partition format — automatically compressed and optimized. Customers cannot directly access the underlying files.
- Storage is built on top of the cloud provider's blob storage: Amazon S3 (AWS), Azure Blob Storage (Azure), or Google Cloud Storage (GCP).
- Billing is separate from compute: you pay for storage even when no warehouses are running.
- Snowflake manages encryption, redundancy, and durability automatically.
- Data is reorganized into micro-partitions (50–500 MB compressed) with rich metadata about each partition (min/max values, distinct count, NULL count) to enable partition pruning during queries.
What the storage layer holds:
- Table data (micro-partitions)
- Time Travel versions of changed data
- Fail-Safe copies
- Stage data (for internal stages)
- Query result cache
Layer 2 — Compute Layer (Virtual Warehouses)
The compute layer consists of Virtual Warehouses (VWs) — named clusters of compute resources that execute SQL queries and DML operations.
Key characteristics:
- Each Virtual Warehouse is an MPP (Massively Parallel Processing) compute cluster consisting of one or more nodes.
- Multiple warehouses can read from the same storage simultaneously — with no resource contention between them.
- Warehouses can be suspended (stopping billing instantly) and resumed (within seconds).
- Size is expressed in T-shirt sizes: X-Small, Small, Medium, Large, X-Large, 2X-Large, 3X-Large, 4X-Large, 5X-Large, 6X-Large.
- Each size-up doubles the compute resources (and credits per hour).
Credit consumption by warehouse size (Standard):
| Size | Credits/Hour |
|---|---|
| X-Small | 1 |
| Small | 2 |
| Medium | 4 |
| Large | 8 |
| X-Large | 16 |
| 2X-Large | 32 |
| 3X-Large | 64 |
| 4X-Large | 128 |
| 5X-Large | 256 |
| 6X-Large | 512 |
Snowpark-Optimized Warehouses consume more credits than Standard warehouses of the same T-shirt size. Do not confuse the two types on the exam.
What the compute layer does:
- Executes SQL queries (SELECT, DML)
- Loads data (COPY INTO)
- Runs Snowpark code
- Performs transformations
What the compute layer does NOT do:
- Store data permanently
- Run Cloud Services operations (those are free up to 10% of compute credits)
Layer 3 — Cloud Services Layer
This is the brain of Snowflake — a collection of services that coordinate all activity across the platform. It runs on Snowflake-managed infrastructure and is always available even when no virtual warehouses are running.
Cloud Services responsibilities:
| Service | Description |
|---|---|
| Authentication | Validates user identity (passwords, MFA, OAuth, key-pair) |
| Infrastructure Management | Provisions, monitors, and repairs compute resources |
| Metadata Management | Tracks table definitions, statistics, partition metadata |
| Query Parsing & Optimization | Parses SQL, generates and optimizes query execution plans |
| Access Control | Enforces RBAC and DAC policies |
| Transaction Management | Ensures ACID compliance across concurrent operations |
Billing note: Cloud Services usage is free up to 10% of daily compute credits consumed. Usage beyond that threshold is billed separately. This is an important exam detail.
-- Cloud Services are used transparently — for example when you run:
SHOW TABLES IN DATABASE MY_DB;
-- This uses Cloud Services (metadata lookup) with no warehouse neededSeparation of Storage and Compute — Why It Matters
This is the most architecturally significant feature of Snowflake and appears frequently on the exam.
Benefits of separation:
- Independent scaling — scale compute up/down without touching storage
- Cost optimization — suspend compute when idle; storage billing continues at low rates
- Workload isolation — multiple teams run their own warehouses against shared data
- No contention — a large analytics query on WH_ANALYTICS doesn't affect ingestion on WH_INGEST
-- Engineering ingests data on their warehouse
USE WAREHOUSE WH_INGEST;
COPY INTO raw.events FROM @my_stage;
-- Meanwhile, BI queries the same data on their own warehouse
USE WAREHOUSE WH_BI;
SELECT date_trunc('hour', event_time), count(*)
FROM raw.events
GROUP BY 1;
-- No queuing between these teams!Snowflake Editions
The edition determines which features are available and the SLA provided. You must know these for the exam.
| Feature | Standard | Enterprise | Business Critical | Virtual Private Snowflake (VPS) |
|---|---|---|---|---|
| Time Travel (max) | 1 day | 90 days | 90 days | 90 days |
| Multi-cluster Warehouses | ❌ | ✅ | ✅ | ✅ |
| Column-level Security (Masking) | ❌ | ✅ | ✅ | ✅ |
| Row Access Policies | ❌ | ✅ | ✅ | ✅ |
| Search Optimization | ❌ | ✅ | ✅ | ✅ |
| HIPAA Compliance | ❌ | ❌ | ✅ | ✅ |
| PCI DSS Compliance | ❌ | ❌ | ✅ | ✅ |
| Tri-Secret Secure (CMK) | ❌ | ❌ | ✅ | ✅ |
| AWS PrivateLink / Azure PE | ❌ | ❌ | ✅ | ✅ |
| Private Deployment | ❌ | ❌ | ❌ | ✅ |
| SLA | 99.5% | 99.9% | 99.95% | 99.99% |
The exam frequently tests edition gates. Remember: Multi-cluster warehouses, Column Masking, and Row Access Policies all require Enterprise or higher.
Cloud Provider and Region Considerations
Snowflake accounts are deployed in a specific cloud + region combination. This is determined at account creation and cannot be changed.
| Cloud | Example Regions |
|---|---|
| AWS | us-east-1, us-west-2, eu-west-1, ap-southeast-1 |
| Azure | eastus2, westeurope, australiaeast |
| GCP | us-central1, europe-west4, asia-northeast1 |
Key points for the exam:
- A single Snowflake account lives in one cloud and one region
- Data can be replicated across regions and clouds using Database Replication
- Cross-Cloud Business Continuity (CCBC) allows failover to a different cloud provider
- Private connectivity (AWS PrivateLink, Azure Private Endpoints) is available on Business Critical+
Key Terminology for the Exam
| Term | Definition |
|---|---|
| Virtual Warehouse (VW) | Named MPP compute cluster that executes queries |
| Micro-partition | Fundamental storage unit: 50–500 MB compressed, columnar |
| Cloud Services Layer | Intelligence layer: auth, optimization, metadata, access control |
| SaaS | Software as a Service — Snowflake manages all infrastructure |
| MPP | Massively Parallel Processing — queries distributed across nodes |
| Separation of Storage & Compute | Storage and compute scale independently, billed separately |
| Credit | Unit of Snowflake compute consumption |
Practice Questions
Q1. Which Snowflake layer is responsible for query optimization and access control enforcement?
- A) Database Storage Layer
- B) Compute Layer
- C) Cloud Services Layer ✅
- D) Virtual Warehouse Layer
Q2. A company wants to ensure their BI reporting queries never compete with their ETL pipelines. Which Snowflake architectural feature enables this?
- A) Micro-partitions
- B) Separation of storage and compute ✅
- C) The Cloud Services layer
- D) Time Travel
Q3. At which Snowflake edition do Multi-cluster Virtual Warehouses first become available?
- A) Standard
- B) Enterprise ✅
- C) Business Critical
- D) Virtual Private Snowflake
Q4. Cloud Services usage is billed only when it exceeds what percentage of daily compute credits?
- A) 5%
- B) 10% ✅
- C) 15%
- D) 20%
Q5. A Snowflake warehouse is suspended for 4 hours. Which costs continue to accrue during this suspension?
- A) Compute costs only
- B) Both compute and storage costs
- C) Storage costs only ✅
- D) No costs accrue
Q6. Which Snowflake edition is required for HIPAA compliance?
- A) Standard
- B) Enterprise
- C) Business Critical ✅
- D) Virtual Private Snowflake
Q7. Snowflake stores data in which underlying format?
- A) Row-oriented flat files
- B) Parquet files directly readable by customers
- C) Compressed columnar micro-partitions ✅
- D) JSON documents
Key Takeaways for Exam Day:
- Three layers: Cloud Services → Compute (VW) → Storage — each independent and separately billed
- Cloud Services = brain (auth, optimization, metadata) — free up to 10% of compute credits
- Storage billing never stops — even when warehouses are suspended
- Multi-cluster WH, Column Masking, Row Access Policies → Enterprise+
- HIPAA / Tri-Secret Secure → Business Critical+
- VPS = highest isolation, 99.99% SLA, private deployment