Trusted Advisor — Redshift Reserved Node Optimization

TL;DR: For EC2, Savings Plans mostly replaced Reserved Instances. For Redshift there is no Savings Plan — Reserved Nodes remain the only commitment vehicle, and the discount is unusually steep (up to ~75% off at 3-year all-upfront). Trusted Advisor flags clusters running on-demand 24×7 for months and its recommendation is usually right and conservative — the savings routinely run $30K–80K/year per cluster. The one hard rule: migrate node family first, buy the RN second.

The numbers

• Discount ladder: 1-yr no-upfront ~25% → 3-yr no-upfront ~50% → 3-yr all-upfront ~75% (the all-upfront delta is wider for Redshift than most services).
• A 3-yr all-upfront RN on an ra3.4xlarge 24×7 ≈ $8,500/yr saved per node — a 4-node cluster is ~$34K/yr from one decision.
• RNs are region + node-family + size scoped, apply automatically to matching nodes, don't auto-renew, and can't be terminated early (no-upfront is the only mid-term walk-away).
• Not covered: Concurrency Scaling, Redshift Spectrum, and RA3 Managed Storage all bill separately.
• Field examples: a 4-node ra3.4xlarge deferred RNs for 14 months and left ~$86K on the table before buying a 3-yr all-upfront (3 of 4 nodes) for ~$7,200/mo saved; a team that bought DC2 RNs then wanted RA3 got stuck paying the stranded DC2 RN — migration must come first.

Do this

1. Trust the finding on stable clusters — if a cluster has run 24×7 for 30+ days, you're likely leaving 25–75% on the table; the check waits for sustained usage before recommending.
2. Migrate node family before buying — if you're on DC2, move to RA3 (separates storage/compute, often cheaper on its own), run on-demand a month to confirm the new node count, then buy the RN.
3. Cover the baseline, not the peak — reserve ~70–80% of steady-state nodes per (region, family, size) bucket; leave scale-up and overnight-ETL nodes on on-demand.
4. Start 1-year no-upfront if learning the system, then move to 3-year all-upfront (mathematically dominant if cash flow allows) once node count and family are proven stable.
5. Buy in the Redshift console (RNs have their own purchase UI, not Cost Explorer) and set a renewal reminder 30–60 days before expiry.

Gotchas

• No Savings Plan is coming — stop deferring RN purchases waiting for one; it's not on AWS's roadmap.
• Order of operations is everything — a DC2 RN bought "to save now" becomes a 30-month tax on stranded capacity if you migrate to RA3 after.
• RA3 Managed Storage is uncovered — the RN discounts node-hours only; storage is on-demand-priced with no commitment vehicle.
• Can't exit early — a 3-year RN is a 3-year financial commitment, so steady-state certainty matters more here than for looser EC2 Savings Plans.

Skip this if

• The cluster is experimental, seasonal, or might not survive the term — stay on-demand until usage is proven, then commit only to the baseline (see Redshift Reserved Nodes for full strategy).
• The workload is bursty or variable rather than 24×7 — Redshift Serverless fits better, or if it's merely idle on a predictable schedule, Pause and Resume. Parent: Trusted Advisor Cost Optimization; share RNs Org-wide via Consolidated Billing.