DE Wikiguides / cloud-platforms

Cloud Data Platforms

Modern data engineering is increasingly cloud-native. Snowflake, Google BigQuery, Amazon Redshift, and Databricks are the major platforms that power analytics at scale. Each offers different trade-offs in compute/storage separation, pricing, and ecosystem.

Snowflake

Snowflake is a fully managed cloud data warehouse with complete separation of compute and storage. It runs on AWS, Azure, and GCP.

Key Features

  • Virtual warehouses - Independent compute clusters that can be resized, paused, or auto-scaled
  • Zero-copy cloning - Instant, storage-free copies of databases for dev/test
  • Time travel - Query and restore data up to 90 days in the past
  • Data sharing - Share live data across Snowflake accounts without copying
  • Snowpipe - Serverless, continuous data ingestion from cloud storage

Pricing Model

Snowflake charges separately for storage (compressed, per TB/month) and compute (per warehouse-hour based on warehouse size). Compute can be paused to zero cost.

-- Create a virtual warehouse
CREATE WAREHOUSE analytics_wh
  WAREHOUSE_SIZE = 'MEDIUM'      -- XSMALL through 6X-LARGE
  AUTO_SUSPEND = 300             -- seconds of inactivity
  AUTO_RESUME = TRUE
  INITIALLY_SUSPENDED = TRUE;

-- Clone a database for development
CREATE DATABASE dev_analytics
  CLONE prod_analytics;

-- Query historical data (time travel)
SELECT *
FROM orders AT(TIMESTAMP => '2025-01-15 12:00:00'::TIMESTAMP);

Google BigQuery

BigQuery is a serverless data warehouse on GCP. It automatically manages compute resources, scaling transparently to petabytes.

Key Features

  • Serverless - No clusters to manage; compute scales automatically
  • Columnar storage - Capacitor columnar format with high compression
  • BigLake - Query external data in GCS, S3, or Azure Blob with unified access control
  • Omni - Run BigQuery queries across multi-cloud (AWS, Azure) from a single interface
  • Partitioning + clustering - Fine-grained data organization for cost and performance
-- Create a partitioned, clustered table CREATE TABLE my_dataset.events ( event_id     STRING, event_name   STRING, user_id      INT64, event_time   TIMESTAMP, payload      JSON ) PARTITION BY DATE(event_time) CLUSTER BY user_id OPTIONS( partition_expiration_days = 365, require_partition_filter = TRUE );

Amazon Redshift

Redshift is AWS's fully managed data warehouse, known for its columnar storage and massively parallel processing (MPP) architecture.

Key Features

  • RA3 nodes - Separate compute and storage with managed storage (Redshift Managed Storage)
  • AQUA - Advanced Query Accelerator for faster scans on compressed data
  • Spectrum - Query data directly in S3 without loading into Redshift
  • Auto WLM - Automatic workload management for concurrent queries

Databricks

Databricks provides a unified analytics platform built on Apache Spark, offering both data engineering and data science workflows.

Key Features

  • Delta Lake - ACID transactions, schema enforcement, and time travel on data lakes
  • Unity Catalog - Fine-grained governance across workspaces and clouds
  • Serverless SQL - Auto-scaling SQL warehouses with per-second billing
  • MLflow - Machine learning lifecycle management built in
  • Delta Sharing - Open protocol for sharing data across platforms

Platform Comparison

FeatureSnowflakeBigQueryRedshiftDatabricks
ServerlessYes (auto-resume)FullyRA3 reduces opsServerless SQL
Compute/storage separationFullFullRA3 nodesVia Delta Lake
Open formatsProprietaryProprietaryProprietaryDelta, Parquet, Iceberg
Multi-cloudAWS, Azure, GCPGCP + OmniAWS onlyAWS, Azure, GCP
Python/notebooksSnowparkBigQuery DataFramesLimitedNative

Resources

Cloud Data Platforms

Modern data engineering is increasingly cloud-native. Snowflake, Google BigQuery, Amazon Redshift, and Databricks are the major platforms that power analytics at scale. Each offers different trade-offs in compute/storage separation, pricing, and ecosystem.

Snowflake

Snowflake is a fully managed cloud data warehouse with complete separation of compute and storage. It runs on AWS, Azure, and GCP.

Key Features

  • Virtual warehouses - Independent compute clusters that can be resized, paused, or auto-scaled
  • Zero-copy cloning - Instant, storage-free copies of databases for dev/test
  • Time travel - Query and restore data up to 90 days in the past
  • Data sharing - Share live data across Snowflake accounts without copying
  • Snowpipe - Serverless, continuous data ingestion from cloud storage

Pricing Model

Snowflake charges separately for storage (compressed, per TB/month) and compute (per warehouse-hour based on warehouse size). Compute can be paused to zero cost.

-- Create a virtual warehouse
CREATE WAREHOUSE analytics_wh
  WAREHOUSE_SIZE = 'MEDIUM'      -- XSMALL through 6X-LARGE
  AUTO_SUSPEND = 300             -- seconds of inactivity
  AUTO_RESUME = TRUE
  INITIALLY_SUSPENDED = TRUE;

-- Clone a database for development
CREATE DATABASE dev_analytics
  CLONE prod_analytics;

-- Query historical data (time travel)
SELECT *
FROM orders AT(TIMESTAMP => '2025-01-15 12:00:00'::TIMESTAMP);

Google BigQuery

BigQuery is a serverless data warehouse on GCP. It automatically manages compute resources, scaling transparently to petabytes.

Key Features

  • Serverless - No clusters to manage; compute scales automatically
  • Columnar storage - Capacitor columnar format with high compression
  • BigLake - Query external data in GCS, S3, or Azure Blob with unified access control
  • Omni - Run BigQuery queries across multi-cloud (AWS, Azure) from a single interface
  • Partitioning + clustering - Fine-grained data organization for cost and performance
-- Create a partitioned, clustered table CREATE TABLE my_dataset.events ( event_id     STRING, event_name   STRING, user_id      INT64, event_time   TIMESTAMP, payload      JSON ) PARTITION BY DATE(event_time) CLUSTER BY user_id OPTIONS( partition_expiration_days = 365, require_partition_filter = TRUE );

Amazon Redshift

Redshift is AWS's fully managed data warehouse, known for its columnar storage and massively parallel processing (MPP) architecture.

Key Features

  • RA3 nodes - Separate compute and storage with managed storage (Redshift Managed Storage)
  • AQUA - Advanced Query Accelerator for faster scans on compressed data
  • Spectrum - Query data directly in S3 without loading into Redshift
  • Auto WLM - Automatic workload management for concurrent queries

Databricks

Databricks provides a unified analytics platform built on Apache Spark, offering both data engineering and data science workflows.

Key Features

  • Delta Lake - ACID transactions, schema enforcement, and time travel on data lakes
  • Unity Catalog - Fine-grained governance across workspaces and clouds
  • Serverless SQL - Auto-scaling SQL warehouses with per-second billing
  • MLflow - Machine learning lifecycle management built in
  • Delta Sharing - Open protocol for sharing data across platforms

Platform Comparison

FeatureSnowflakeBigQueryRedshiftDatabricks
ServerlessYes (auto-resume)FullyRA3 reduces opsServerless SQL
Compute/storage separationFullFullRA3 nodesVia Delta Lake
Open formatsProprietaryProprietaryProprietaryDelta, Parquet, Iceberg
Multi-cloudAWS, Azure, GCPGCP + OmniAWS onlyAWS, Azure, GCP
Python/notebooksSnowparkBigQuery DataFramesLimitedNative

Resources