AWS Analytics Service Selection

Choose from the data path, not from a service list

AWS analytics services solve different stages of a data problem. A useful starting point is to ask whether data arrives continuously or in batches, where the durable source of truth lives, whether transformation is required before analysis, what query interface users expect, and whether the final consumer is an engineer, an analyst, or a dashboard reader. The answers usually identify a pipeline rather than a single service.

For example, Kinesis Data Streams is designed to capture and expose ordered streams of records for near-real-time consumers. AWS Glue can discover metadata and run ETL. Athena can query data in Amazon S3 and federated sources with SQL without provisioning a traditional database server. EMR runs open-source distributed frameworks such as Spark and Hadoop. Redshift is a managed data warehouse optimized for analytical SQL, while Amazon Quick provides the business-intelligence presentation layer.

Data lake, processing engine, and warehouse are different roles

A data lake commonly keeps files in Amazon S3 and separates durable storage from the engines that inspect or transform them. The AWS Glue Data Catalog can hold table and schema metadata for that data without moving the underlying objects. Athena can then query cataloged data directly, and EMR or Glue can perform larger transformations when a SQL-only query is not the right processing model.

A data warehouse such as Amazon Redshift is a different boundary. It provides a purpose-built analytical database with columnar storage, distributed execution, workload optimization, and SQL semantics. Redshift Spectrum can extend warehouse queries to data in S3, so lake and warehouse architectures can coexist. The choice is therefore not simply S3 or Redshift; it is about where data is stored, where compute executes, and what performance and management model a workload needs.

• Athena: ad-hoc/serverless SQL over data in S3 and supported federated sources.
• Glue: catalog, discovery, and managed data integration/ETL.
• EMR: managed execution of open-source big-data frameworks and custom distributed processing.
• Redshift: analytical data warehousing, with Spectrum for external S3 data.
• Quick: dashboards, analyses, and BI consumption rather than upstream ETL.

Streaming and BI complete the end-to-end picture

When records arrive continuously and multiple applications must process them with low delay, Kinesis Data Streams provides a shard-based stream abstraction. Partition keys determine how records are assigned across shards, and consumers can process records independently. A streaming stage can feed transformations or persistent destinations before the resulting data is queried with services such as Athena or Redshift.

Business intelligence sits at the opposite end of the data path. Amazon Quick connects to datasets and produces interactive analyses and dashboards for people who need to explore or consume results. It does not replace the storage, catalog, or processing layer. Keeping these boundaries clear prevents a common architecture mistake: selecting a BI tool for data engineering work or a processing engine when the requirement is simply serverless SQL over existing files.