Domain 4 › Topic 6
⭐⭐⭐⭐⭐ Huge Conceptual Area

⚡ Performance Optimization

Performance optimization is about making your application faster, cheaper, and more scalable. It covers reducing latency, tuning memory and compute, using caches effectively, and choosing the right architecture patterns.

🔑 Covers: Latency Reduction · Memory Tuning · Caching · Read Scaling · Profiling · Cost Optimization

🎯 The Five Goals of Performance Optimization

⚡ Reduce Latency

Make responses faster. Users expect API responses in <500ms. Use caching, reduce hops, move compute closer to data.

📈 Improve Scalability

Handle 10x more users without rewriting code. Use serverless, auto-scaling, horizontal scaling, event-driven architectures.

💾 Optimize Memory

Right-size memory allocation. Too little = out of memory errors. Too much = wasted cost. Use profiling to find the sweet spot.

💰 Reduce Cost

Eliminate waste. Right-size resources. Use spot instances, reserved capacity. Turn off what's not needed.

🔧 Improve Reliability

Remove single points of failure. Use queues, retries, circuit breakers, multi-AZ deployments.

🔄 Read Scaling

For read-heavy workloads: add read replicas (RDS), use caching (ElastiCache/DAX), use CDN (CloudFront).

🔑 Caching — The Most Powerful Optimization

Caching stores the result of an expensive operation (like a database query) so future requests can get the answer instantly without doing the expensive operation again. It's the single most impactful optimization in most applications.

Where to Add Caching in AWS

Caching Layers in AWS Architecture
🌐CloudFrontCDN Edge Cache — closest to user
↓ cache miss ↓
🚦API Gateway CacheCache API responses at gateway level
↓ cache miss ↓
λLambdaIn-memory cache in execution env (module-level variables)
↓ cache miss ↓
🚀ElastiCacheDistributed in-memory cache — milliseconds
⚡DynamoDB DAXMicrosecond reads for DynamoDB
↓ cache miss ↓
🗄️Database (RDS/DynamoDB)Source of truth — milliseconds to seconds

⏱️ Cache TTL (Time-to-Live)

TTL (Time-to-Live)
How long a cached item is valid before it must be refreshed from the source. Short TTL = fresher data but more cache misses. Long TTL = stale data risk but better cache hit rate. Choose based on how often the underlying data changes. Example: Product prices change daily → TTL = 1 hour. User profile data changes rarely → TTL = 24 hours.
🎯 Caching in Exam Scenarios
S3 getting throttled (503 SlowDown errors): Add CloudFront in front of S3. CloudFront caches S3 responses at the edge.
RDS database overloaded with reads: Add ElastiCache (Redis or Memcached) in front of RDS.
DynamoDB reads too slow: Add DynamoDB DAX (microsecond read latency).
API Gateway backend slow for repeated identical requests: Enable API Gateway caching.

📊 Profiling — Find What to Optimize

Profiling means measuring which parts of your code consume the most resources. You can't optimize effectively if you don't know where the time and memory are actually being spent. Guessing is inefficient — 80% of time is usually spent in 20% of the code.

🔧 AWS Profiling Tools

🤖 Amazon CodeGuru Profiler

Continuously profiles your running application in production. Shows flame graphs and identifies expensive code lines. See Topic 7: CodeGuru for full details.

🔍 AWS X-Ray

Shows which service in a multi-service request takes longest. Identifies subsegments (individual DB calls, HTTP calls) that are slow. Great for distributed system profiling.

💾 Memory Optimization

For Lambda especially, memory directly controls compute power (more memory = more CPU). Finding the right memory size requires testing, not guessing:

  • Too little memory: Function runs slowly, may timeout, may run out of memory
  • Right amount: Fast enough, cost-efficient
  • Too much memory: You pay for unused capacity
  • Lambda Power Tuning: Automated tool that tests multiple memory configurations and finds the optimal cost/performance balance (see Topic 8)

🔄 Read Scaling Patterns

When your application is read-heavy (more reads than writes), scale reads independently:

ServiceRead Scaling OptionBenefit
RDSRead Replicas (up to 15)Distribute read queries across multiple DB instances
RDSElastiCache in frontCache frequent reads in memory — millisecond response
DynamoDBDynamoDB DAXMicrosecond read latency, handles millions of reads/sec
S3CloudFront distributionCache S3 objects at 400+ edge locations globally
API GatewayResponse cachingCache API responses for configurable TTL
💡 Optimization Mindset
The optimization order: (1) Measure first — use CloudWatch, X-Ray, CodeGuru to find the actual bottleneck. (2) Optimize the bottleneck — don't optimize parts that aren't slow. (3) Measure again — verify the optimization worked. (4) Consider cost — faster is not always worth more expensive.
← AWS CloudTrail