Ch7. ELB and Auto Scaling — Building Highly Available Architectures

Elastic Load Balancing (ELB)

Distributes incoming traffic across multiple backend servers.

ALB (Application Load Balancer)

Layer 7 (HTTP/HTTPS) load balancer.

Advanced routing:

Path-based: /api/* → API servers, /static/* → S3
Host-based: api.example.com → API tier, www.example.com → web tier
HTTP header and query string routing
WebSocket and HTTP/2 support

Best for: Web applications, REST APIs, microservices.

NLB (Network Load Balancer)

Layer 4 (TCP/UDP) load balancer.

Ultra-low latency, extreme performance
Static IP or Elastic IP support
Handles millions of requests per second

Best for: Gaming servers, IoT, financial trading, real-time streaming.

Exam decision tree: Need HTTP routing rules? → ALB. Need ultra-low latency TCP/UDP or a static IP? → NLB. Need to integrate with third-party firewalls/IDS/IPS? → GLB.

GLB (Gateway Load Balancer)

Layer 3 load balancer. Integrates with third-party network appliances (firewalls, IDS/IPS).

Auto Scaling Group (ASG)

Components

Launch Template: AMI, instance type, security groups, IAM role
Group size: Minimum / Maximum / Desired capacity
Scaling policies

Scaling Policy Types

Dynamic scaling:

Simple: CloudWatch alarm → add/remove a fixed number
Step: Different scaling steps for different alarm thresholds
Target Tracking: Maintain a target metric (e.g., keep CPU at 50%)

Scheduled scaling: Pre-scale for known patterns (e.g., scale to 10 instances every weekday at 8 AM).

Predictive scaling: ML analyzes historical patterns → scales proactively before demand arrives.

Health Checks

EC2 health check: Hardware/software status of the instance
ELB health check: Application-level response (recommended — more granular)

Unhealthy instances are automatically terminated and replaced.

HA Architecture Pattern

Standard multi-tier HA:

Internet → ALB (public subnets, multi-AZ)
              → ASG (private subnets, multi-AZ)
                   → RDS Multi-AZ (private subnets)

Key Concept Cards

ALB vs. NLB ★★★★★ : ALB=Layer 7 HTTP routing, NLB=Layer 4 high-performance TCP. HTTP web app=ALB, gaming/real-time=NLB.

Target Tracking Scaling ★★★★★ : Set a target (e.g., CPU 50%) and ASG automatically adjusts instance count. Simplest and most recommended approach.

ELB Health Checks ★★★★☆ : More granular than EC2 health checks. Validates application-level responses (/health endpoint) to replace truly unhealthy instances.

Practice Quiz

Q1. A microservices app needs /api/v1, /api/v2, and /web to route to different server groups. What load balancer?

ALB (Application Load Balancer). ALB path-based routing rules can direct /api/v1 to Target Group 1, /api/v2 to Target Group 2, and /web to Target Group 3. NLB operates at Layer 4 and cannot inspect URL paths.

Q2. Traffic predictably spikes 10x between 9 AM and 6 PM. What scaling approach is optimal?

Combine Scheduled Scaling with Target Tracking. Use Scheduled Scaling to pre-scale capacity before 9 AM (before load arrives), then let Target Tracking handle real-time adjustments during the day. Predictive Scaling is also valid if the pattern repeats consistently across weeks.