Scalability and Performance Engineering
ELB is fully managed and automatically scales. However, architecture still matters.
Throughput and Latency
- ALB handles HTTP-level routing logic.
- NLB provides lower latency and higher throughput.
If you need millions of requests per second with minimal overhead, NLB is typically preferred.
Connection Handling
Load balancers manage:
- TCP connection reuse
- Idle timeouts
- Keep-alive settings
Improper configuration can cause:
- Connection exhaustion
- Uneven load distribution
You must tune idle timeout settings based on application behavior.
Cross-Zone Load Balancing
When enabled:
- Traffic is evenly distributed across all targets.
- Prevents imbalanced AZ load.
In production, uneven traffic distribution often leads to “hot” zones. Cross-zone balancing mitigates that.
Observability
CloudWatch metrics provide:
- Request count
- Target response time
- HTTP error rates
- Healthy host count
Production systems should:
- Alert on 5xx spikes
- Alert on unhealthy targets
- Track latency percentiles
Load balancers become observability checkpoints in distributed systems.
In this section, I learned:
0 of 4 completed