Storage tiering separates the FHIR data that clinicians touch every hour from the data that lives in the archive for compliance. The two workloads have different access patterns, different latency tolerances, and different cost per gigabyte. A tiering strategy that respects the split can cut long-term storage cost by an order of magnitude without changing anything a clinician notices.
The trick is to draw the tier boundaries where the access patterns actually live. For related walkthroughs, related FHIR tooling explainers collects the surrounding material.
What Hot Data Actually Looks Like
Hot FHIR data is the current-state slice that clinical workflows read constantly. Current Patients, active Encounters, recent Observations, open MedicationRequests. The read latency requirement is milliseconds; the availability requirement is very high; the cost per gigabyte is highest.
Hot storage typically lives on the primary FHIR server's persistence tier. A pass through the site's FHIR storage bill projector can bracket the hot-tier footprint against the patient count.
What Warm Data Looks Like
Warm FHIR data is the recently-inactive slice: Encounters closed six to twelve months ago, Observations older than the current window, MedicationRequests fulfilled and archived. Warm reads happen but are rare enough that seconds of latency are acceptable.
Warm storage often lives on a read replica of the primary tier or on a separate storage engine with lower per-gigabyte cost. The trade is longer read latency for cheaper writes.
Cold Data Is the Compliance Archive
Cold FHIR data is the historical archive: version history older than the retention rolling window, closed episodes older than a few years, attachments older than a defined threshold. Cold reads happen for audit or clinical explainer flows and are tolerated at minutes.
Cold storage lives on object storage or archival tiers with per-gigabyte cost an order of magnitude below warm. For the history-specific slice, history table vs current state: doubling your storage bill quietly covers the shape.
Drawing the Tier Boundaries
The boundaries should match the access pattern, not the calendar. A workflow that revisits a Patient after two years wants that Patient on hot storage regardless of the calendar age. A workflow that never touches an Observation after six months can migrate it to warm at that boundary.
Two common patterns work:
- Calendar-based tiering, moving data at fixed ages. Simple to implement, easy to explain, sometimes wasteful.
- Access-based tiering, moving data after N days of no reads. More complex to implement, better cost outcomes, harder to explain to auditors.
Most deployments start with calendar-based tiering and add access-based rules where the savings justify the complexity.
Rehydration Cost Matters
Every tiering strategy carries a rehydration cost: pulling cold data back into a warm or hot tier when a workflow needs it. A well-designed strategy accepts occasional rehydration and monitors its frequency. A poorly-designed one rehydrates constantly and gives up most of the cost saving.
Report rehydration as a first-class metric. Deployments that ignore it discover it as a line item on the invoice.
Retention Interacts With Tiering
Retention policies decide what data expires; tiering decides where non-expired data lives. The two are independent but interact. A deployment that never expires data still benefits from tiering; a deployment that expires aggressively still benefits from tiering the retained data.
For the retention framing, how retention policies shape long-term FHIR storage cost is the accompanying reference.
The Sizing Implication
A tiered storage strategy shifts what the sizing plan looks like. Instead of one storage number, the plan carries hot, warm, and cold as separate lines, each with its own cost per gigabyte and its own growth curve. For the up-front estimation approach, estimating storage before you commit to a FHIR platform is the accompanying reference.
Tiering is one of the highest-leverage storage decisions a FHIR deployment makes. Drawing the boundaries deliberately pays back every quarter of the deployment's life.
Sources
- HL7 FHIR Bulk Data Access IG covering archive-style access - HL7 FHIR Bulk Data Access IG covering archive-style access patterns adjacent to cold tiering