`
Most engineers assume load balancer costs scale with throughput: "We push 2 Gbps, so our bill should reflect 2 Gbps."
That intuition is wrong.
AWS doesn't bill ALB or NLB by throughput. They bill by LCU (Load Balancer Capacity Units)—and you pay for the single largest dimension, not the sum.
The Rule: Pay for the Max, Not the Sum
Four dimensions are evaluated every hour:
- New connections/flows per second
- Active connections/flows
- Bandwidth (Gbps)
- Rule evaluations per second (ALB only, beyond 10 free rules)
$$\text{LCU} = \lceil \max(\text{dim}_1, \text{dim}_2, \text{dim}_3, \text{dim}_4) \rceil$$
Because AWS bills only for the maximum dimension, optimizing the other three saves exactly $0.
When the Economics Break
1. The High-Connection Workload (38× Price Gap)
- ALB: 25 new connections/sec per LCU
- NLB: 800 new flows/sec per LCU (32× more efficient)
At 50,000 new connections/sec:
- ALB requires 2,000 LCU → ~$11,712/month
- NLB requires 63 LCU → ~$308/month
2. The Silent Rule-Evaluation Tax
Every rule beyond the first 10 evaluates against incoming requests:
$$\text{LCU}_{\text{rules}} = \frac{(\text{rules} - 10) \times \text{rps}}{1000}$$
At 5,000 rps with 50 path-based routing rules, rule evaluation alone forces 200 LCU ($1,168/mo)—even if you're pushing negligible bandwidth.
Interactive Calculator & Full Breakdown
You can model your own traffic profile and identify your binding dimension with our zero-data-collection, client-side calculator:
👉 Run the ALB vs NLB Cost Calculator on jslet.com`
Top comments (0)