Skip to content

Architecting storage around the access patterns it must support

Choosing the right storage architecture on Google Cloud depends on how data will be accessed: by query pattern (analytical vs transactional), latency requirements, data structure, and access frequency. The Professional Data Engineer exam expects you to map access patterns, OLTP, OLAP, real-time serving, batch analytics, file-based ML training, to the correct storage service and schema design.

1 · Learn the must-know

  • Cloud SQL and AlloyDB suit low-latency, transactional (OLTP) access with strong consistency and relational joins, while BigQuery is optimized for large-scale analytical (OLAP) scans, not row-level lookups.
  • Bigtable is designed for high-throughput, low-latency key-based access patterns (time series, IoT, ad serving) and requires careful row-key design to avoid hotspotting and support range scans.
  • Spanner supports globally distributed, strongly consistent transactional workloads that need horizontal scale beyond a single Cloud SQL instance, using interleaved tables and choice of primary keys to avoid hotspots.
  • Cloud Storage is the default choice for unstructured/semi-structured data and staging (data lakes), with storage classes (Standard, Nearline, Coldline, Archive) chosen based on access frequency to optimize cost.
  • Firestore/Datastore fit document-style, hierarchical access patterns for mobile/web apps needing flexible schema and offline sync, whereas Memorystore serves ultra-low-latency caching for frequently accessed data.
  • A common gotcha: picking BigQuery for high-frequency single-row transactional updates (bad fit due to cost/latency) or Bigtable for ad hoc SQL analytics (lacks native SQL joins) indicates a mismatch between access pattern and storage engine.

2 · Check your understanding

Check this objectiveFree · always available

A logistics company ingests telemetry from 2 million delivery vehicles, writing a GPS coordinate and speed reading for each vehicle every second. Analysts need sub-10-millisecond reads of a single vehicle's most recent 500 readings, and the write throughput must scale linearly as the fleet grows. Which storage architecture best supports this access pattern?

Your objective map0 tried · 0 answered correctly · 67 untouched

What you have tried across GCP PDE's objectives, not a readiness score.

Designing data processing systems~22% of the exam0 of 17 tried
Ingesting and processing the data~25% of the exam0 of 11 tried
Storing the data~20% of the exam0 of 13 tried
Preparing and using data for analysis~15% of the exam0 of 11 tried
Maintaining and automating data workloads~18% of the exam0 of 15 tried

3 · Keep going