Caching and Performance Bottleneck Detection

What it is

A cache keeps a copy of data that is expensive to fetch or compute in a faster place, so repeat requests skip the slow path. Caching cuts latency and reduces load on the database or backend.

A bottleneck is the one component that limits overall speed. You cannot tune what you have not found, so detection comes first, then caching and other fixes.

How it works

Choosing the cache

Amazon ElastiCache

ElastiCache offers Valkey, Redis OSS and Memcached engines. Valkey and Redis OSS add replication, high availability, persistence and rich data structures. Memcached is simpler and has no replication.

Your code decides how the cache is filled:

• Lazy loading (cache aside): read the cache first; on a miss, read the database and write the result to the cache. Only requested data is cached, but a miss costs extra time and data can go stale.
• Write through: update the cache whenever you write to the database. Data stays fresh, but you cache items that may never be read.
• TTL: every entry gets an expiry, which bounds staleness.

DynamoDB Accelerator

DAX is a managed in memory cache that is API compatible with DynamoDB. You point the DAX client at the cluster endpoint instead of using the DynamoDB client.

• Eventually consistent reads (the default) are served from the cache on a hit.
• Strongly consistent reads pass through to DynamoDB and are not cached.
• The item cache holds GetItem and BatchGetItem results, and the query cache holds Query and Scan result sets. Both default to a five minute TTL.

Use DAX for a repeated read heavy workload against DynamoDB. It does not help write heavy tables or workloads that need strongly consistent reads.

Amazon CloudFront

CloudFront caches responses at edge locations. The cache lifetime comes from the cache policy's minimum, default and maximum TTL, and from the origin's Cache-Control (max-age, s-maxage) or Expires headers. Without a cache policy the default TTL is 24 hours.

• To update an object before it expires, create an invalidation, or better, use versioned file names such as app.v2.js so the new name is a cache miss.
• The cache key decides what makes two requests different. Including extra headers, cookies or query strings lowers the hit rate, so include only what the origin needs.

API Gateway caching

Caching is enabled per stage on REST APIs. The default TTL is 300 seconds and the maximum is 3,600; a TTL of 0 disables it. By default only GET methods are cached. Choose which request parameters form the cache key. Cache hits and misses show as CacheHitCount and CacheMissCount in CloudWatch.

Detecting bottlenecks

Read the pattern: rising Duration with flat traffic points at a slow dependency, Throttles point at a limit, and a few keys dominating DynamoDB throttling points at a hot partition key.

AI tools for optimization

• CloudWatch investigations: analyzes telemetry for an issue and suggests related metrics, logs and likely causes.
• Kiro and Amazon Q Developer: chat in the IDE can review a function, point out repeated calls inside loops or work that could be cached, and suggest refactors. The Q Developer IDE plugins end on April 30, 2027, and Kiro replaces them.
• AWS Compute Optimizer: recommends Lambda memory settings from observed usage.

These produce suggestions. Test each one and compare metrics before and after the change.

How the exam asks about this

• "A read heavy DynamoDB table needs microsecond latency with minimal code change" -> put DAX in front of the table.
• "Users in other Regions load static assets slowly" -> serve them through CloudFront with a suitable TTL.
• "Which tool shows which downstream call adds the most latency to a request?" -> the X-Ray service map and trace timeline.