ALB vs NLB

AWS LBs.

Overview

ALB vs NLB is the choice between Application Load Balancer (HTTP/HTTPS-aware, layer 7, host and path routing) and Network Load Balancer (TCP/UDP-direct, layer 4, ultra-low latency). Both have their place; the right answer depends on the protocol and the operational fit, not on which one the team used last quarter.

• ALB: layer 7, HTTP-aware. Host-based routing, path-based routing, target groups by URL pattern. Right for HTTP and HTTPS workloads.
• NLB: layer 4, TCP/UDP-direct. Static IPs, ultra-low latency, preserves source IP. Right for non-HTTP protocols and latency-sensitive paths.
• Protocol per workload. Match the LB to what the workload actually speaks. HTTPS goes to ALB; gRPC over TCP often goes to NLB.
• Operational fit per team. Existing tooling and dashboards bias the choice. Document the rationale rather than re-deriving it per service.

The approach

Workload-driven choice, per-protocol fit, documented rationale. The discipline is making the LB choice once per workload with a written reason rather than relying on whoever provisioned it most recently.

• Workload-driven. LB choice per workload. Reality drives the answer, not team preference.
• ALB for HTTP-aware. HTTP and HTTPS workloads with host or path routing requirements. Default for web traffic.
• NLB for TCP-direct. TCP and UDP workloads, latency-sensitive paths, static-IP requirements. Default for non-HTTP.
• Documented choice plus operational fit. Per-LB rationale captured; team workflow fit considered. Future investigation has the breadcrumb.

Why this compounds

Each correctly matched LB produces ongoing operational value. Wrong-LB workloads pay latency or cost penalties indefinitely; right-LB workloads pay neither. By year two the team's LB choice is automatic and the documentation explains why so newcomers can audit it.

• Better operational fit. LB matches the workload. Velocity stays high.
• Better cost efficiency. Pricing model matches the traffic profile. NLB cheaper for high-volume TCP, ALB richer for HTTP.
• Workload-driven decisions. Replaces tribal preference with documented rationale. Quality of LB choice improves.
• Year-one investment, year-two habit. First LB choice is the investment; subsequent ones run on the same rationale.