Cloud Technology and Services: Choosing Compute, Storage, Database, and Network
This is the largest domain. Most questions are of the form "given this workload description, which service fits?" - so learn the decision boundaries between services rather than their feature lists.
Compute
EC2 for full control of the OS and long-running servers. Lambda for event-driven code with no server management, billed per request and duration, with a 15-minute maximum execution time. ECS/EKS with Fargate for containers without managing instances. Elastic Beanstalk for deploying applications while AWS handles capacity. Lightsail for simple predictable workloads with a flat monthly price. Instance families follow a pattern: general purpose (M/T), compute optimized (C), memory optimized (R/X), storage optimized (I/D), accelerated (P/G).
Storage
S3 is object storage, effectively unlimited, accessed by API - it is not a mountable file system. Classes trade retrieval speed for price: Standard, Intelligent-Tiering for unknown or changing access, Standard-IA and One Zone-IA for infrequent access, Glacier Instant Retrieval (milliseconds), Glacier Flexible Retrieval (minutes to about 12 hours), and Glacier Deep Archive (about 12 hours standard, longer for bulk). EBS is block storage attached to one instance within one Availability Zone. EFS is shared, elastic file storage across a Region. FSx covers Windows and high-performance file workloads. Storage Gateway bridges on-premises systems to AWS storage.
Databases and analytics
RDS/Aurora for relational workloads; Aurora is AWS-built with higher performance and availability. DynamoDB for key-value at any scale with single-digit-millisecond latency. Redshift for data warehousing; Athena for querying S3 with SQL and no infrastructure; ElastiCache for in-memory caching; Kinesis for streaming data.
Network and delivery
A VPC holds subnets, route tables, and gateways; Route 53 is DNS; CloudFront is the CDN that caches content at edge locations; Global Accelerator routes traffic over the AWS backbone without caching; Direct Connect is a private dedicated link; API Gateway fronts APIs.
Worked example
A compliance archive must be retained seven years and is retrieved only during audits, with 12 hours of retrieval tolerance acceptable. S3 Glacier Deep Archive fits: lowest storage cost, retrieval measured in hours. If the auditors needed the file in seconds, Glacier Instant Retrieval would be correct; if retrieval times varied wildly, Intelligent-Tiering would automate the choice.
Traps that cost marks
- Picking Lambda for a job that runs longer than 15 minutes.
- Treating S3 as a shared file system for an application that expects a mounted drive - that is EFS.
- Choosing Redshift for occasional ad-hoc queries over S3 (Athena) or DynamoDB for relational joins.
- Forgetting that an EBS volume cannot be attached across Availability Zones, while EFS is regional.