Mansi
HOMEWORKASKABOUTCONTACT
HOMEWORKASKABOUTCONTACT
Resume

Home/Work

Project 04

From OLAP to Workload-Specific Data Architecture

A workload fork — one stressed warehouse becomes two specialized systems.

Lead Data Engineer · Data Platforms · Cloud Architecture · ETL ModernizationData Architecture · Cloud · Performance Engineering

AWSRedshift ServerlessS3 TablesAurora PostgreSQLDynamoDBAthena / Presto
Workload splitMixed Redshift access forks into serving (DynamoDB / Aurora) and analytics (S3 Tables / Athena).SharedServingAnalyticsFitApplicationAPI / backendRedshiftmixed workloadDynamoDBpoint lookupsAuroraoperational CRUDS3 Tablesbulk / lakeAthenaaggregationEngine ↔ pattern
Mixed Redshift access forks into serving (DynamoDB / Aurora) and analytics (S3 Tables / Athena).

01 — The problem

An analytical warehouse was also serving high-frequency application lookups; continuous queries prevented effective scale-down and raised cost.

02 — The system

Select a layer to see its role.

Data flow

  1. 01ServingHigh-frequency application point lookups
  2. 02AnalyticsBulk and aggregation workloads

Technology — why it was used

AWS

Redshift Serverless

S3 Tables

Aurora PostgreSQL

DynamoDB

Athena / Presto

05 — Evidence

Application serving

  • Point reads
  • High frequency
  • User latency
→

Analytics workload

  • Scans
  • Aggregations
  • BI
Match the engine to the access pattern — one warehouse was doing two jobs.

Takeaway

Match the engine to the access pattern — not every workload belongs in the warehouse.

  • Reframe 'replace the platform' into 'separate the workloads' when evidence shows mismatch.

What landed

  • Reframed the problem from replacing Redshift to separating workloads
  • Identified hybrid direction: dedicated serving layer plus S3-based analytics
PreviousWeb Intelligence PipelineAll workHomeBack to the map
Mansi

© 2026 Mansi Dhruv. All rights reserved.

Created with curiosity