Amazon Redshift Storage: Architecture, Storage Types, RA3 Nodes, and Managed Storage

Amazon Redshift is AWS's fully managed, petabyte-scale cloud data warehouse designed for high-performance analytics. One of the key factors behind Redshift's performance is its storage architecture, which has evolved significantly from traditional local-storage-based nodes to the modern RA3 architecture with Redshift Managed Storage (RMS).

Amazon Redshift Architecture Overview

A Redshift cluster consists of:

• Leader Node

The leader node receives SQL queries from client applications, creates execution plans, and coordinates query execution across compute nodes.

Compute Nodes

Compute nodes execute query operations and store data. Data is distributed across nodes using a massively parallel processing (MPP) architecture for high-performance analytics.

Storage Layer

Redshift stores data in a columnar format and uses compression techniques to reduce storage consumption and improve query performance. Depending on the node type, storage may reside on local SSDs, HDDs, or Redshift Managed Storage backed by Amazon S3.

Evolution of Redshift Storage

1. Dense Storage (DS2) Nodes

DS2 nodes use traditional HDD-based storage.

Characteristics

• Large storage capacity

• Lower cost per TB

• Suitable for large datasets with less demanding performance requirements

• Legacy architecture

Limitations

• Storage and compute are tightly coupled

• Scaling storage requires adding more compute nodes

• Slower than SSD-based architectures

AWS generally recommends migrating DS2 workloads to RA3 for better performance and scalability.

2. Dense Compute (DC2) Nodes

DC2 nodes use local NVMe SSD storage.

Characteristics

• High-performance SSD storage

• Excellent for small to medium-sized datasets

• Fast query execution

Limitations

• Storage and compute remain tightly coupled

• Scaling storage requires additional nodes

• Less flexible for rapidly growing data warehouses

DC2 is often recommended for smaller data warehouses where datasets remain relatively stable.

3. RA3 Nodes (Current Generation)

RA3 represents the third-generation Redshift architecture and introduces a major innovation: separation of compute and storage.

Instead of sizing a cluster based on both storage and compute requirements, organizations can provision compute independently and pay only for the managed storage they actually use.

RA3 Architecture

                    Client Applications
                             │
                             ▼
                     +---------------+
                     | Leader Node   |
                     +---------------+
                             │
          ┌──────────────────┴──────────────────┐
          ▼                                     ▼

  +---------------+                     +---------------+

  | Compute Node  |                     | Compute Node  |

  | Local SSD     |                     | Local SSD     |

  +---------------+                     +---------------+

          │                                     │
          └──────────────┬──────────────────────┘
                         ▼
            Redshift Managed Storage (RMS)
                         │
                         ▼
                     Amazon S3

The local SSDs act as a high-speed cache, while Amazon S3 provides virtually unlimited durable storage behind the scenes.

What is Redshift Managed Storage (RMS)?

Redshift Managed Storage automatically manages where data resides.

Hot Data

Frequently accessed data blocks are stored on high-performance local SSDs for low-latency access.

Cold Data

Less frequently accessed data is automatically moved to Amazon S3. Redshift intelligently retrieves the data when required.

Intelligent Data Placement

Redshift continuously analyzes:

• Data temperature (access frequency)

• Data age

• Query patterns

• Workload characteristics

Based on these factors, it automatically moves data between SSD cache and S3 storage without administrator intervention.

Key Benefits of RA3 and Managed Storage

Independent Scaling

Compute and storage can be scaled separately.

Benefits:

• Lower infrastructure costs

• Better resource utilization

• Simplified capacity planning

Automatic Tiering

No manual management of hot and cold data is required.

Administrators do not need to create custom storage tiers or archive strategies.

Cost Optimization

Organizations pay separately for:

• Compute resources

• Managed storage consumption

This model is particularly beneficial for large datasets with moderate query workloads.

High Performance

RA3 uses:

• High-bandwidth networking

• Local SSD caching

• Intelligent prefetching

• Automatic data eviction

to provide performance similar to local storage while maintaining cloud-scale capacity.

Available RA3 Node Types

The appropriate node type depends on workload complexity, concurrency requirements, and data volume.

RA3 vs DC2 vs DS2

When Should You Choose RA3?

RA3 is ideal when:

• Data volume exceeds several terabytes

• Storage growth is unpredictable

• Compute and storage requirements vary independently

• You want simplified operations and lower management overhead

• You require petabyte-scale analytics

AWS generally recommends RA3 for most new Redshift deployments due to its flexibility and managed storage architecture.

Conclusion

Amazon Redshift's storage architecture has evolved from tightly coupled compute-storage models (DS2 and DC2) to the modern RA3 architecture powered by Redshift Managed Storage. By combining high-performance local SSD caching with Amazon S3-backed storage, RA3 enables organizations to scale storage and compute independently, reduce costs, and simplify administration while maintaining enterprise-grade analytical performance. For most modern data warehouse workloads, RA3 with Managed Storage has become the preferred Redshift deployment model.

Personal Notes: Unlike Amazon Athena, Amazon Redshift physically stores data inside the data warehouse. It is designed specifically for Online Analytical Processing (OLAP) workloads and uses Massively Parallel Processing (MPP) along with columnar storage to achieve high query performance.

1. Stores Data

What does it mean?

Amazon Redshift is a database, meaning it physically stores your tables and data.

For example:

CREATE TABLE sales (

   sale_id INT,

   customer_id INT,

   amount DECIMAL(10,2),

   sale_date DATE

);

When data is loaded,

COPY sales

FROM 's3://company-data/sales/'

IAM_ROLE 'arn:aws:iam::123456789012:role/redshift-role'

FORMAT AS PARQUET;

the data is copied into Redshift's storage.

Unlike Athena,

• Athena → reads data directly from S3.

• Redshift → stores data inside the warehouse.

Internal Architecture

               Client

                  │

                  ▼

         Amazon Redshift

        -------------------

        sales table stored

        customer table stored

        product table stored

        inventory table stored

The data remains inside Redshift until:

• DELETE

• TRUNCATE

• DROP TABLE

• VACUUM (reclaims deleted space)

Interview Point

Redshift is a storage + compute engine. 

Athena is only a query engine.

2. Columnar Storage

If you execute >> SELECT SUM(salary) FROM sales;

Redshift reads only Salary Column instead of the entire row. This reduces disk I/O dramatically.

Advantages

✓ Reads only required columns

✓ Faster aggregation

✓ Better compression

✓ Less disk I/O

✓ Ideal for analytics

Columnar storage is one of the biggest reasons Redshift performs much faster than traditional row-oriented OLTP databases for analytical queries.

3. Compression

Because each column stores similar data types, compression becomes extremely effective.

Why compression is effective

Since every column contains similar values, patterns are easier to compress than in row storage.

Benefits

✓ Less storage

✓ Faster disk reads

✓ Better cache utilization

✓ Reduced network transfer

Compression Types

Redshift automatically selects the best encoding.

Common encodings include:

• AZ64

• LZO

• ZSTD

• BYTEDICT

• DELTA

• RUNLENGTH

Most modern Redshift clusters use Automatic Table Optimization, which chooses encodings automatically.