The 4 Azure Load Balancers
Azure offers four distinct load-balancing services. Each operates at a different layer and scope — understanding their boundaries is the single most important decision for AZ-305.
| Service | Layer | Scope | Key Features | Cost Tier | Best For |
|---|---|---|---|---|---|
| Azure Load Balancer | L4 (TCP/UDP) | Regional | HA ports, health probes, SNAT, zone-redundant | Low (per-rule + data) | Non-HTTP workloads, internal tier routing |
| Application Gateway | L7 (HTTP/S) | Regional | Path/host routing, WAF v2, SSL offload, autoscale, private | Medium (capacity units) | Web apps needing WAF, path-based routing within a region |
| Azure Front Door | L7 (HTTP/S) | Global | Anycast, WAF, caching, SSL offload, instant failover, Rules Engine | Higher (per request + data transfer) | Global web/API with low latency, DDoS + WAF at edge |
| Traffic Manager | DNS | Global | Priority, weighted, geographic, performance, subnet routing | Low (per million queries) | Non-HTTP global routing, DR failover, DNS-level distribution |
🔑 Core Distinction
- L4 sees only IP + port. Fastest, but no URL awareness.
- L7 inspects HTTP headers, URL paths, cookies. Can route /api/* differently from /static/*.
- DNS-based (Traffic Manager) never touches data traffic — it resolves a name to an endpoint IP. TTL governs failover speed (30s–300s).
- Anycast (Front Door) routes the TCP connection itself to the nearest POP — sub-second failover, no DNS TTL wait.
Decision Tree: Choosing the Right Load Balancer
Use this flowchart on exam day and in real designs:
🔑 Quick Decision Rules
- Non-HTTP traffic? → Load Balancer (regional) or Traffic Manager (global DNS failover).
- HTTP + single region? → Application Gateway (especially if WAF or path routing needed).
- HTTP + multi-region + performance? → Front Door (anycast, instant failover, edge caching).
- Multi-region + non-HTTP or legacy? → Traffic Manager (DNS-based, protocol agnostic).
Azure Front Door Deep Dive
Front Door is Azure's premium global L7 load balancer. It uses anycast — every Front Door POP advertises the same IP, so clients connect to the nearest POP automatically via BGP routing.
🔑 Front Door Capabilities
- Anycast routing: User connects to nearest of 180+ edge POPs. No DNS wait for failover — TCP itself is re-routed.
- Instant failover: If a backend region fails health probes, traffic shifts to next-best backend in <30 seconds — no TTL delay.
- WAF at the edge: Block attacks at the POP before traffic reaches your origin. Managed rule sets (OWASP 3.2, bot protection).
- Caching: Static content cached at edge POPs — reduces origin load and latency.
- SSL offload: Front Door terminates TLS, forwarding HTTP or re-encrypted HTTPS to backends.
- Rules Engine: URL rewrite, redirect, header manipulation, rate limiting — all at the edge.
- Private Link origins: Front Door Premium can connect to private backends (no public IP on origin).
Front Door Tiers
- Standard: CDN + basic routing + managed WAF rules. Good for content delivery.
- Premium: Everything in Standard + Private Link origins + enhanced WAF (bot protection, geo-filtering) + Microsoft Threat Intelligence rules.
Traffic Manager: DNS-Based Global Routing
Traffic Manager works at the DNS layer. It never proxies traffic — it simply resolves your FQDN to the best endpoint IP based on the routing method you configure.
🔑 Routing Methods
- Priority: Active/passive failover. All traffic to endpoint #1 unless it fails health check → endpoint #2.
- Weighted: Distribute traffic by percentage (e.g., 80/20 for canary deployments).
- Performance: Route to the lowest-latency endpoint based on the user's DNS resolver location.
- Geographic: Route based on the geographic origin of the DNS query (e.g., EU users → EU endpoint). Useful for data sovereignty.
- Subnet: Map specific client IP ranges to specific endpoints.
- MultiValue: Return multiple healthy endpoints in DNS response (client picks one).
When Traffic Manager is Sufficient
- Non-HTTP protocols (SQL, RDP, custom TCP)
- You don't need edge WAF or caching
- You can tolerate DNS TTL-based failover (30–300 seconds)
- You need geographic routing for compliance (e.g., "German users must only resolve to the Frankfurt endpoint")
- Budget-conscious global routing — Traffic Manager is the cheapest global option
Application Gateway: Regional L7 Powerhouse
App Gateway is your in-region HTTP load balancer with deep L7 features. From AZ-104 you know the basics — here's the architect's view.
🔑 Architect-Level Decisions
- Path-based routing: /api/* → API backend pool, /images/* → storage, /app/* → App Service.
- Multi-site hosting: Route by Host header — consolidate multiple apps behind one gateway.
- WAF v2: OWASP Core Rule Set 3.2, custom rules, exclusions. Required for PCI-DSS workloads in-region.
- Autoscaling (v2): Scales from 0 to 125 capacity units based on traffic. No pre-provisioning needed.
- Private-only: Deploy in a private subnet with no public IP — for internal apps or behind Front Door.
- Mutual TLS (mTLS): Client certificate authentication at the gateway.
- Connection draining: Gracefully remove backend instances during deployments.
v1 vs v2 SKU
Always use v2 for new deployments. v1 is legacy — no autoscale, no zone redundancy, no key vault integration. The exam assumes v2.
Combining Load Balancers
Enterprise architectures often layer multiple load-balancing services. Common patterns:
Pattern 1: Front Door + Application Gateway
- Front Door handles global routing, edge WAF, caching, SSL offload.
- App Gateway (private) in each region handles path-based routing, additional WAF rules, and routes to backend compute (AKS, VMs, App Service).
- Lock down App Gateway to only accept traffic from Front Door (check
X-Azure-FDIDheader or use Private Link with Front Door Premium).
Pattern 2: Traffic Manager + Load Balancer
- Traffic Manager resolves DNS to the healthiest region.
- Load Balancer (L4) in each region distributes TCP/UDP across backend VMs.
- Use case: Non-HTTP workloads needing global failover (gaming servers, IoT ingestion, database replicas).
Pattern 3: Front Door + Load Balancer (internal)
- Front Door terminates HTTP at edge, sends to App Service or VMs.
- Internal Load Balancer distributes traffic between application tier and data tier within a VNet.
Cross-Region HA Patterns
🔑 Active-Active vs Active-Passive
- Active-Active: Both regions serve traffic simultaneously. Front Door (performance routing) or Traffic Manager (weighted/performance). Requires data replication strategy.
- Active-Passive: Secondary region is standby. Traffic Manager (priority) or Front Door with failover routing. Simpler data consistency but higher RTO.
Design Considerations
- Health probes: All services probe backends — configure probe paths that actually validate app health (not just TCP port open).
- Data replication: The load balancer handles traffic routing, but you must separately ensure data is available in the failover region (Cosmos DB multi-region, SQL Geo-Replication, Storage GRS).
- Failback strategy: After primary recovers, how do you shift traffic back? Automatic (risky) or manual (slower but safer)?
- Cost: Active-active doubles compute cost. Use consumption-based services (Functions, App Service auto-scale to 0) to reduce standby costs in passive regions.
Real-World: Global SaaS with Regional Compliance
📋 Scenario
A SaaS company serves customers in US, EU, and APAC. Requirements:
- EU customer data must stay in EU (GDPR)
- Sub-100ms latency for all regions
- Survive a full region failure with <1 min failover
- WAF protection at the edge
- Path-based routing: /api/* to microservices, /app/* to frontend
Solution Architecture
- Azure Front Door Premium — global entry point with anycast, edge WAF, Private Link to origins.
- Geographic routing rules in Front Door — EU users pinned to EU backend group (GDPR compliance); US/APAC routed by latency.
- Application Gateway (private, WAF v2) in each region — path-based routing to AKS clusters.
- Internal Load Balancer within each region's VNet for service-to-service traffic between microservices.
- Cosmos DB with multi-region writes — EU region as primary write for EU data, with geo-fencing.
Why Not Alternatives?
- Traffic Manager alone? — DNS TTL delay fails the <1 min failover SLA. No edge WAF.
- Front Door without App Gateway? — Works for simple backends, but the team needs private-only backends with in-region WAF layering and mTLS to services.
- Load Balancer for global? — L4 only, regional only. Can't meet global latency or L7 routing needs.