PROVIDER Β· AWS columnar cloud data warehouse
Amazon Redshift
Amazon Redshift is AWS's columnar MPP cloud warehouse where RA3 separates storage from compute, Spectrum queries the S3 lake directly, and Glue, QuickSight, and SageMaker complete the AWS analytics stack.
4.4pts
Pros
Cons
Product Information
Detailed Notes
Overview
Amazon Redshift launched in 2012 as AWS's earliest managed warehouse on the ParAccel columnar MPP architecture, adding RA3 for separated storage and compute in 2019 and a Serverless form in 2021, long the default warehousing choice on AWS.
For Backend & Database work the advantage is seamless integration with the AWS data stack, since data sits in S3 queried through Spectrum external tables without loading, COPY loads in parallel from S3 at high speed, and Glue, EMR, QuickSight, and SageMaker complete an analytics platform.
Accept capacity planning reality, because RA3 bills per node and needs sizing previews with elastic resize still incurring waits, and compared with serverless options operations remain tunable through distribution keys, sort keys, VACUUM, and WLM queues, where poor choices visibly underperform.
Core Strengths
- Columnar MPP architecture keeps large aggregations and joins consistently fast
- Spectrum queries the S3 data lake directly without loading
- RA3 separates storage and compute with cross-cluster data sharing
- Native integration with Glue, EMR, QuickSight, and SageMaker
Product Ecosystem
RA3 provisioned clusters
Billed per node type and count with elastic resize and managed storage.
Redshift Serverless
Billed on RPU usage with automatic start and stop for intermittent load.
Spectrum and data sharing
External tables query Parquet, ORC, or CSV in S3, and data sharing allows cross-account and cross-cluster reads.
Drawbacks
- Provisioned clusters need sizing previews and undersizing visibly hurts performance
- Distribution keys, sort keys, and VACUUM require expertise with poor settings degrading queries
- AWS-only, so cross-cloud migration rebuilds pipelines and permissions
- High-concurrency BI needs extra concurrency scaling configuration
Use Cases
- Data warehousing on AWS (β β β β β ) - Seamless with S3 and Glue
- Direct analysis of the S3 lake (β β β β β ) - Spectrum without loading
- Intermittent analytical load (β β β β ) - Serverless starts and stops
- Cloud-neutral warehousing (β β ) - AWS bound
Pricing
| Plan | Price | Scope |
|---|---|---|
| Serverless | Per RPU-hour | Auto start and stop for variable load |
| RA3 provisioned | Per node-hour | Managed storage billed separately |
| Spectrum scans | Per TB scanned | External tables over S3 |
Control cost by separating storage from compute, since RA3 managed storage is cheap while compute bills per node or RPU, making Serverless cheapest for intermittent load and provisioned clusters with reserved instances better for steady heavy use, with Spectrum scans likewise bounded.
FAQ
-
Redshift or Athena? Redshift for recurring high-frequency BI queries and scheduled reports, Athena for occasional exploration directly over S3 at lower cost.
-
Provisioned or Serverless? Provisioned with reserved instances for steady predictable load, Serverless for a few hours daily or highly variable traffic.
-
Why are queries slow? Usually poor distribution and sort key design, stale statistics, or missing VACUUM, so reviewing plans and keys often beats adding nodes.