Services Comparison

Quick reference for choosing the right AWS service.

Endpoint Types

Type	Latency	Payload	Timeout	Cost Model	Use Case
Real-time	Milliseconds	6 MB	60s	Per hour	Interactive apps
Serverless	Seconds (cold start)	6 MB	60s	Per request	Variable traffic
Async	Minutes	1 GB	15 min	Per hour	Large payloads
Batch Transform	Hours	-	-	Per job	Offline batch

Data Processing Services

Service	Mode	Use Case	Scalability
Glue ETL	Batch	Large-scale ETL	Auto-scaling
Glue DataBrew	Batch	Visual prep	Managed
EMR	Batch	Custom Spark	Manual/Auto
Kinesis Data Streams	Streaming	Custom consumers	Shards
Kinesis Firehose	Streaming	Direct to S3/etc	Automatic
SageMaker Processing	Batch	ML data prep	Instance-based

Storage for ML

Service	Data Type	Access Pattern	Use Case
S3	Any	Object access	Data lake, models
Feature Store Online	Features	Low-latency	Real-time inference
Feature Store Offline	Features	Batch queries	Training
Redshift	Structured	SQL queries	Analytics
DynamoDB	Key-value	Low-latency	Application data

MLOps Orchestration

Service	Focus	Best For
SageMaker Pipelines	ML-specific	End-to-end ML workflows
Step Functions	General	Multi-service orchestration
MWAA (Airflow)	General	Complex DAGs, existing Airflow
EventBridge	Events	Event-driven triggers

Hyperparameter Tuning Strategies

Strategy	Best For	Speed	Quality
Bayesian	Most cases	Fast	High
Random	Exploration	Medium	Medium
Grid	Small space	Slow	Complete
Hyperband	Deep learning	Fast	High

Monitoring Services

Service	Purpose	Data Type
Model Monitor	Drift detection	ML metrics
CloudWatch Metrics	Infrastructure	Time series
CloudWatch Logs	Application logs	Text
CloudTrail	API audit	Events
X-Ray	Distributed tracing	Traces

Security Services

Service	Purpose
IAM	Identity and access
KMS	Encryption keys
Secrets Manager	Credentials
Macie	Sensitive data discovery
VPC	Network isolation
Security Groups	Instance firewall

AI Services vs Custom ML

Use Case	AI Service	Custom (SageMaker)
Text sentiment	Comprehend	Custom NLP
Object detection	Rekognition	Custom CV
Speech-to-text	Transcribe	Custom ASR
Translation	Translate	Custom NMT
Recommendations	Personalize	Custom RecSys
Forecasting	Forecast	Custom models

Bedrock vs SageMaker

Aspect	Bedrock	SageMaker
Model type	Foundation models	Any ML model
Training	Fine-tuning only	Full training
Infrastructure	Serverless	Managed instances
Customization	Limited	Full control
Use case	GenAI apps	Custom ML