EC2 Instance Family Cost

Right family per workload.

Overview

EC2 instance family cost chooses the right family for each workload. Default to general-purpose and the bill is fine and the performance is mediocre; matching the family to the workload pattern (CPU-bound, memory-bound, IO-bound) produces real cost-performance.

• Right family per workload. Per-workload family selection; the workload pattern determines the family, not vendor recommendation.
• Per-family cost-performance. Per-family cost-performance ratio; the headline price hides the cost-per-unit-of-work delta.
• Compute-optimized (c). Per-CPU-bound c-family; matches workloads where vCPU is the constraint.
• Memory-optimized (r) plus quarterly review. Per-memory-bound r-family for high-memory workloads; quarterly family review catches drift as workload evolves.

The approach

The practical approach: per-workload family selection, c-family for CPU-bound, r-family for memory-bound, quarterly family review, documented per-service rationale. The team’s discipline produces matched compute instead of generic over-provisioning.

• Per-workload family. Per-workload family selection; the family matches the workload constraint.
• Compute-optimized for CPU-bound. Per-CPU-bound c-family; the higher CPU-to-RAM ratio matches the workload.
• Memory-optimized for memory-bound. Per-memory-bound r-family; high-memory workloads run cheaper on r-family than scaled-up general-purpose.
• Per-quarter family review plus documented rationale. Quarterly family review catches drift; per-service family rationale committed for review.

Why this compounds

Family discipline compounds across services. Each correct family produces ongoing cost-performance; the team’s compute discipline grows; new services inherit the family-selection methodology.

• Better cost efficiency. Right family matches workload; the bill tracks the workload constraint, not generic over-provisioning.
• Better performance. Right family matches workload; the workload runs at design capacity instead of fighting the wrong resource ratio.
• Better operational fit. Right family matches team; the operational pattern stays simple when families match consistently.
• Institutional knowledge. Each decision teaches compute patterns; the team’s capacity engineering muscle grows.

Family discipline is an operational discipline that pays off across years. Nova AI Ops integrates with compute telemetry, surfaces patterns, and supports the team’s compute discipline.