Panduan optimasi instance reserved

Reserved Instances Optimization Guide

Reserved Instances (AWS), Reservations (Azure), and Committed Use Discounts (GCP) are the single biggest lever for cutting the biaya of steady-state cloud beban kerjas. A 1-year commitment typically saves 30 to 40 percent off on-demand, and a 3-year commitment saves 50 to 72 percent — savings that, for a beban kerja spending $10,000 per month, can free up $36,000 to $86,000 per year.

What Is a Reserved Instance?

A Reserved Instance (RI) is not a separate instance type — it is a billing diskon applied to on-demand usage that matches the reservation’s specifications. You continue to launch the same VM you always have; the cloud provider’s billing system applies the diskoned rate to the matching hours of usage. Unused reservations do not roll over, so the goal is to size the commitment to your baseline usage, not your peak.

The Three Commitment Horizons

HorizonAWS discountAzure discountGCP discountBest for
1-year No Upfront~40% off~40% off~30% off (CUD)Steady-state workloads under 12 months old
1-year All Upfront~45% off~42% off~37% offWorkloads with predictable spend and cash on hand
3-year All Upfront~60% off~62% off~57% off (CUD)Multi-year workloads with stable architecture

The 1-year No Upfront option is the safest starting point: no cash outlay, and the commitment can be exchanged within the instance family if your beban kerja changes. Move to 3-year All Upfront only for beban kerjas you are certain will run unchanged for the full term, since the break-even point is typically 12 to 15 months.

How to Stagger Reservations

A common mistake is to commit the entire fleet on the same start date, which forces an all-or-nothing renewal 12 or 36 months later. Instead, stagger reservations:

  1. Month 1 — commit 30 percent of baseline usage as a 1-year reservation.
  2. Month 2 — commit another 30 percent, again 1-year.
  3. Month 3 — commit the final 30 to 40 percent.

After 12 months you are renewing one-third of your reservations each month rather than the whole fleet, which gives you room to right-size, switch instance families, or migrate providers without stranding committed spend.

Savings Plans vs Reserved Instances

On AWS and Azure, the modern alternative to RI is a Savings Plan (AWS) or Savings Plan (Azure). A Savings Plan commits to a dollar-per-hour spend rather than a specific instance family, so the diskon applies across any instance size or family in the committed region. The diskon is slightly lower (up to 66 percent on AWS Compute Savings Plans vs 72 percent on RIs), but the flexibility is worth it for teams whose instance mix changes over time.

GCP Committed Use Discounts

GCP works differently. Sustained-Use Discounts (SUDs) apply automatically when a VM runs more than 25 percent of the month, with the diskon scaling up to 30 percent off at 100 percent utilization — no commitment required. Committed Use Discounts (CUDs) stack on top of SUDs for an additional 21 to 57 percent off, and GCP’s CUDs are unique in that they apply to vCPU and memory independently, so you can commit CPU without committing memory.

Common Mistakes to Avoid

  • Over-committing — committing more than your baseline leaves you paying for unused hours. Always commit to the 30-day average, not the peak.
  • Ignoring convertible RIs — AWS Convertible RIs biaya slightly more but can be exchanged for a different instance family, region, or operating system at no penalty.
  • Forgetting RI expiration — set calendar reminders 30 days before each reservation expires, so you can renew or downsize before the diskon lapses.
  • Mixing spot and RI for the same instance — if part of your beban kerja runs on spot, do not also reserve the same instance type; the RI diskon is wasted when the instance is on spot.

See our AWS vs Azure compute perbandingan for the specific per-instance reservation savings, and the biaya optimization panduans overview for the next layer of savings.

Other languages