Design High-Performing Architectures

Where the Time Actually Goes

Performance answers depend on identifying the bottleneck: distance, repeated work, the wrong store, or the wrong compute shape. Each has its own set of correct answers and its own tempting wrong ones.

Exam tip. Two options often both work. Pick the one built for the exact scenario - that is what the question is testing.

Picking the Right Database Before Tuning the Wrong One

AWS deliberately offers a purpose-built store per data shape, and the exam rewards recognising the shape in the scenario rather than defaulting to relational.

Exam tip. When a scenario names a data shape, the purpose-built store almost always beats the general-purpose one, even if both could technically work.

Why Scaling Sometimes Does Not Help

Several performance questions describe a system that scales and still fails. The cause is almost always a bottleneck that scaling cannot move.

Exam tip. When a scenario says scaling is already working and it still fails, the answer is a bottleneck downstream of the thing that scaled.

1. Which ElastiCache engine should be chosen when the cache must support replication, persistence and failover?

• A. Either, since both support replication
• B. Neither, a cache cannot be replicated
• C. Rediscorrect
• D. Memcached

Redis supports replication, snapshots, persistence, and automatic failover, along with richer data structures. Memcached is a simpler multi-threaded cache that scales horizontally but has no replication or persistence.

2. A global audience is downloading large static assets from an S3 bucket in one Region and users far from that Region report slow downloads. What is the most effective change?

• A. Enable S3 Versioning on the bucket
• B. Increase the size of the EC2 instances serving the site
• C. Serve the assets through Amazon CloudFront so they are cached at edge locationscorrect
• D. Move the bucket to a larger storage class

CloudFront caches content at edge locations close to users, cutting latency and origin load for repeated downloads. Storage class affects cost and retrieval characteristics rather than distance, versioning tracks object history, and the bottleneck here is geography rather than compute.