SAA-C03 Study summary

Design High-Performing Architectures

Design High-Performing Architectures

saa-c03Design High-Performing ArchitecturesAWS
Last organized

Match compute, storage, database, networking, and caching choices to the workload’s access patterns and bottlenecks.

In one sentence

Match compute, storage, database, networking, and caching choices to the workload’s access patterns and bottlenecks.

Key points

  • Scale horizontally when possible and use managed or serverless services where their scaling model fits.
  • Choose storage by object, block, or file access and databases by relational, key-value, document, graph, or cache patterns.
  • Use CloudFront, read replicas, DAX, or ElastiCache only when their cache layer matches the data and consistency needs.

Exam takeaway

Measure latency, throughput, concurrency, and cost; do not optimize from service names alone.

Analytics & AI/ML performance

Performance or architecture needStart with
Ad hoc SQL directly over S3Athena + partitions / columnar formats / compression
Sustained, frequent, complex analytics and BIAmazon Redshift
Real-time streams, multiple consumers, custom processingKinesis Data Streams
Low-operations near-real-time delivery to S3 and other targetsAmazon Data Firehose
Serverless ETL and data catalogingAWS Glue
Large-scale custom Spark or Hadoop processingAmazon EMR
Train and deploy models with proprietary dataAmazon SageMaker AI
Build generative-AI apps through managed foundation-model APIsAmazon Bedrock

Design reminders

  • Isolate production DynamoDB traffic from large historical analytics and training workloads.
  • Athena scan volume depends on partitioning, compression, and file format.
  • Firehose buffers records, so verify whether its delivery latency meets the requirement.
  • Training and inference are different workloads; plan compute, scaling, and monitoring separately.
  • When model quality drops, check data quality, schema, and training-data freshness first.

Region, edge, and scope

  • Choose a Region by compliance, latency, service availability, and price.
  • Edge Locations / PoPs bring CloudFront, Route 53, and Global Accelerator closer to users but are not ordinary EC2 deployment locations.
  • CloudFront caches HTTP/HTTPS content, Route 53 returns DNS answers, and Global Accelerator optimizes TCP/UDP network paths.
  • Regional services and zonal resources have different placement scopes.
  • Service availability in a Region is not the same as high availability.