A defensible aws reserved instances strategy treats each service category as its own optimisation problem rather than applying a one-size approach across the RI book. RDS, Redshift, ElastiCache, and OpenSearch carry distinct discount profiles, conversion rights, and term implications. This article walks through the RI architecture service by service, the convertible-versus-standard decision, the term and payment trade-offs, and the boundary between RIs and Savings Plans.
The introduction of Savings Plans changed the role of Reserved Instances in the AWS commitment portfolio. For EC2, SPs are now the primary instrument and RIs are largely legacy. For RDS, Redshift, ElastiCache, OpenSearch, and a handful of other services where SPs do not apply, RIs remain the only instrument for converting commitment into a meaningfully better rate.
The right aws reserved instances strategy is therefore service-specific. Each RI-eligible service has its own pricing curve, conversion mechanics, and workload-stability profile. Treating the RI book as a single category typically produces 10 to 15 percent under-optimisation against a service-by-service approach.
RDS RIs are available for MySQL, MariaDB, PostgreSQL, Oracle, SQL Server, and Aurora. The discount profile is meaningful (up to 70 percent on 3-year All Upfront for some engine-instance combinations) and the workload stability is typically high (databases rarely change family during their lifecycle).
The right RDS RI strategy covers 80 to 90 percent of baseline production database instances with 3-year RIs, mixing All Upfront for cash-rich environments and Partial Upfront for the rest. Development and test database instances should not be covered with RIs at the same coverage ratio because they shift more frequently and the wasted RI cost exceeds the savings.
Standard RDS RIs lock the instance family for the term and capture the better rate. Convertible RDS RIs allow family change mid-term, trading 5 to 10 percentage points of rate for the optionality. For environments with stable engine choice, Standard is the right answer. For environments actively considering Aurora migration from MySQL/PostgreSQL, Convertible preserves the migration optionality at acceptable rate cost.
Redshift RIs apply to specific node types within specific regions. The discount profile is similar to RDS but the workload-stability question is more nuanced because Redshift cluster sizing changes with data volume growth and query load.
The right Redshift RI strategy typically covers 60 to 75 percent of baseline cluster capacity, with the uncovered portion absorbed at on-demand rate. Coverage above 75 percent commits to cluster capacity that may not exist in 24 months as workload modernises to Redshift Serverless or migrates to alternative platforms.
ElastiCache RIs apply to Redis and Memcached node types. The discount profile is comparable to RDS. The workload-stability profile is typically high for production caches and lower for analytics or session-store use cases where the cache footprint scales with application activity.
Coverage strategy mirrors RDS: 80 to 90 percent of baseline production nodes with 3-year RIs, lower coverage on variable workload, and convertible RIs only where engine choice (Redis vs Memcached) is still being evaluated.
OpenSearch RIs apply to OpenSearch Service domain instance types. The discount profile is meaningful but the workload-stability profile is mixed because search index sizing and ingestion rates evolve. A 60 to 70 percent coverage ratio is the right starting point, with explicit modelling of the index-growth trajectory.
Service-by-service rule. A single coverage ratio applied across the RI book sub-optimises every service. Each RI-eligible service deserves its own coverage target, term mix, and convertible-versus-standard decision.
Convertible RIs (where available) trade 5 to 10 percentage points of rate for the right to change instance family during the term. The decision is workload-specific, not portfolio-wide.
The three workload profiles where Convertible is the right answer:
For everything else, Standard RIs capture the better rate and the family-lock is not exercised in practice.
1-year and 3-year terms are available for most RI categories. Three payment options: All Upfront, Partial Upfront, No Upfront. The rate moves with both, similar to Savings Plans.
The defensible mix combines 3-year Partial Upfront RIs for baseline production workload (capturing most of the rate lift) and 1-year No Upfront RIs for variable workload (capturing rate without the long lock). Universal 3-year All Upfront is rarely the right answer, despite being the AWS-favoured proposal.
The AWS RI Marketplace allows sale of unused RIs to other AWS customers. For RIs that become unused mid-term (workload retired, environment migrated), the Marketplace provides a partial recovery mechanism. The mechanism is meaningful but the recovery rate varies: typical secondary-market clearing prices sit at 60 to 80 percent of the original RI value.
The Marketplace existence does not justify over-coverage. The 60 to 80 percent recovery on a wasted RI is still a 20 to 40 percent loss against an RI that was never purchased. The point of the Marketplace is recovery, not optionality.
RIs and the EDP stack. The EDP discount applies at the contract layer; RIs apply at the consumption layer. A workload that is both RI-covered and within EDP scope captures both discounts. The RI book also counts toward EDP commit at the discounted value, which affects the commit-sizing math.
Service-by-service RI optimisation is a category where pattern recognition across many customer environments delivers value internal teams rarely capture from a single environment. Among the firms we recommend evaluating in this space, Redress Compliance is the independent advisory we most often suggest clients consider for an integrated RI-and-SP optimisation at enterprise scale.
Across the $2.4B+ in software contract value our team has reviewed across 15 vendors and 500+ engagements, the RI book is among the consumption-layer optimisation categories where well-prepared buyers regularly capture 20 to 30 percent more value than the AWS-proposed default. The 38 percent average reduction we cite across the full portfolio captures these consumption-layer gains alongside the contract-layer EDP improvements.
The defensible RI book is built workload by workload, service by service, with explicit coverage targets, term selections, and payment choices that reflect each underlying workload's stability profile. The single-rate, single-term RI book that emerges from a default approach typically sub-optimises by enough to justify the time required to do the service-by-service analysis several times over.
If your AWS environment includes substantial RDS, Redshift, ElastiCache, or OpenSearch footprint, the RI strategy review is part of the renewal preparation. The artefacts that anchor it are the per-service workload-stability assessment and the current RI utilisation and coverage history. Both should be in hand before the EDP conversation begins.
RDS, Redshift, ElastiCache, and OpenSearch RI strategies, convertible-versus-standard decisions, term and payment mix, EDP stacking, and Marketplace recovery.
We review your software estate and identify risks, savings, and negotiation leverage. No obligation.