Reserved Instance Strategy

S-003 FinOps · Cloud Cost Optimization Module · Updated 2025-04-28

01 What Is a Reserved Instance? [ CONCEPT ]

// Overview

A Reserved Instance (RI) or Reserved Capacity commitment lets you pay upfront (full or partial) or commit to 1–3 years of usage in exchange for discounts of 30–72% vs on-demand pricing. This is the single most impactful commitment decision in cloud cost optimization.

Rule of thumb: If a workload runs ≥ 8 hours/day, 365 days/year, a 1-yr RI almost always pays for itself within 3–4 months.

02 1-Year vs 3-Year — The Math [ ANALYSIS ]

Commitment	Upfront	Typical Discount	Break-even	Best For
1-yr, No Upfront	$0	~30–40%	3–6 months	Uncertain demand, early-stage workloads
1-yr, Partial Upfront	~30–40% of total	~40–50%	1–3 months	Stable baseline with some variance
1-yr, All Upfront	100% paid	~50–55%	Immediate	Completely predictable steady-state
3-yr, All Upfront	100% paid (3yr)	~60–72%	1–4 months	Long-lived, stable, core infrastructure
3-yr, Partial Upfront	~50–60% of total	~55–65%	2–5 months	Balance of savings and flexibility

⚠ 3-Year risk: Technology moves fast. A 3-yr commitment on an instance family that becomes obsolete (e.g., m5 → m6i migration) creates stranded capacity. Always verify the instance family's expected lifecycle before committing 3 years.

03 Savings Plans vs Reserved Instances [ COMPARISON ]

Feature	EC2 Savings Plans	Convertible RIs	Standard RIs
Discount	Up to 72%	Up to 54%	Up to 72%
Instance family flexibility	Flexible (compute)	Convertible with upgrade	Fixed family/size
AZ flexibility	Region-wide	Same AZ only	Same AZ only
Instance size flexibility	Normalized family	Limited	Fixed size
Commitment type	Dollar amount (compute)	Specific instance type	Specific instance type
Best fit	Variable, mixed workloads	Partial flexibility needed	Fixed, predictable steady-state

Recommendation: For most teams, a blend of Compute Savings Plans (flexible, region-wide) for variable workloads + Standard RIs for your true steady-state baseline delivers the best risk-adjusted savings.

04 Sizing Your Commitment — Safe Floor Method [ PROCESS ]

// Step-by-step sizing process

# Step 1: Baseline (30-day lookback)
aws ce get-rightsizing-recommendations \
  --service "AmazonEC2" \
  --start-date $(date -d '30 days ago' +%Y-%m-%d) \
  --end-date $(date +%Y-%m-%d) \
  --granularity DAILY

# Step 2: Find your true minimum (P95 of daily peak usage)
# Only commit up to the P95 — not the average — to avoid over-commitment

# Step 3: RI coverage ratio targets
Steady-state prod DB:   100% → Standard 3-yr All Upfront
Steady-state app servers: 80–90% → Standard 1-yr Partial Upfront
Batch/nightly workloads: 0% → On-demand + Spot
Dev/test (8–5):         ~30% → Compute Savings Plan (flexible)
Variable APIs:            40–60% → Compute Savings Plan

⚠ Never commit more than your P95 usage. Any usage above your commitment still bills at on-demand rates, but unused RI capacity is still fully paid — you lose twice.

05 Coverage Monitoring — Runbook [ OPERATIONS ]

// Monthly RI coverage review checklist

Export on-demand spend from Cost Explorer (last 30 days)
Cross-reference against active RI/Savings Plan inventory
Flag any instance families with >15% unused RI capacity → sell on marketplace or exchange
Flag any on-demand spend that represents >P80 baseline → buy additional coverage
Check for new workloads that have run 30+ days at steady utilization → evaluate RI candidacy
Review upcoming RI expirations (90-day advance warning)
Validate that retired instance types don't leave orphaned RIs

Disclaimer: This guide provides general informational content about cloud infrastructure cost management. Figures and benchmarks are based on publicly available industry averages (e.g., Gartner, IDC, cloud provider documentation) and may vary by provider, region, and workload. This content is not a substitute for professional financial, legal, or technical advice specific to your organization.