tech-studies

Backend Performance

2 min read

Backend Performance

roadmap.sh: https://roadmap.sh/best-practices/backend-performance

Suggested path through the Backend Performance nodes. Each node links to its lesson when written.

Nodes

Foundations

  • Why backend performance matters
  • Latency vs throughput
  • Percentiles (p50 / p95 / p99) over averages
  • Define SLIs, SLOs and SLAs
  • Establish performance budgets
  • The USE method (Utilization, Saturation, Errors)
  • The RED method (Rate, Errors, Duration)

Measure Before Optimizing

  • Profile before you optimize
  • Identify the bottleneck first
  • CPU profiling
  • Memory profiling
  • Flame graphs
  • Distributed tracing
  • Load testing
  • Stress testing
  • Benchmarking methodology
  • Avoid premature optimization

Database

  • Add the right indexes
  • Avoid N+1 queries
  • Analyze query plans (EXPLAIN)
  • Select only needed columns
  • Use connection pooling
  • Batch reads and writes
  • Paginate large result sets
  • Denormalize when justified
  • Read replicas for read-heavy loads
  • Partitioning and sharding
  • Tune isolation levels
  • Keep transactions short
  • Use prepared statements

Caching

  • Cache expensive computations
  • In-memory caching (Redis / Memcached)
  • Application-level caching
  • HTTP caching headers (ETag, Cache-Control)
  • CDN for static and cacheable content
  • Cache invalidation strategy
  • Avoid cache stampede
  • Set sensible TTLs

Application Layer

  • Use async / non-blocking I/O
  • Offload work to background jobs / queues
  • Reuse connections (HTTP keep-alive)
  • Compress responses (gzip / Brotli)
  • Pick efficient serialization formats
  • Stream large responses
  • Choose efficient algorithms and data structures
  • Avoid blocking the event loop
  • Tune thread / worker pools
  • Mind garbage collection pressure

Concurrency & Scaling

  • Horizontal vs vertical scaling
  • Load balancing
  • Statelessness for scale-out
  • Rate limiting and throttling
  • Backpressure
  • Apply the circuit breaker pattern
  • Set timeouts and retries with backoff
  • Use bulkheads to isolate failures

Network & Infrastructure

  • Reduce round trips
  • HTTP/2 and HTTP/3
  • Keep services close to data (locality)
  • Edge computing
  • Right-size compute resources
  • Autoscaling policies

Observability

  • Metrics collection
  • Centralized logging
  • Tracing across services
  • Dashboards and alerting
  • Continuous performance regression testing

Resources

See resources.md.

Project ideas

  • Build a load-testing harness with k6 against a sample API and chart p50/p95/p99 latency as you add indexes and caching.
  • Take a deliberately N+1-ridden ORM endpoint and refactor it (eager loading, batching, indexes), documenting the EXPLAIN plans before and after.
  • Add a Redis caching layer with proper TTLs and stampede protection in front of a slow endpoint, then measure hit rate and tail-latency improvement.

1 item under this folder.