Hot vs Cold Storage

The four tiers

Exact pricing varies by provider and region but the relative shape is consistent.

The economic crossover

The tiering decision is fundamentally: "what's the expected retrieval frequency?" Cheaper storage with retrieval fees wins for rarely-accessed data; more expensive storage wins for frequently-accessed data.

Rough heuristic for cloud storage:

• Read once per day or more often → hot.
• Read once per week to once per month → warm / infrequent access.
• Read a few times per year → cold.
• Read once a decade or never → archive.

Each tier transition point has provider-specific economics; run the math for your access pattern.

What hot storage gets you

• Single-digit-millisecond first-byte latency.
• High IOPS for random access.
• High throughput for sequential reads.
• No retrieval fees beyond standard egress.
• No minimum storage duration.

Used for active application data, working sets, anything users interact with synchronously.

What warm storage gets you

• Same latency as hot — usually milliseconds.
• Lower per-GB storage cost.
• Per-GB retrieval fee on every read.
• Often a minimum storage duration (e.g., 30 days).

Used for backup repositories that may be restored occasionally, analytics datasets queried weekly, logs queried during investigations.

What cold storage gets you

• Much lower storage cost (often 1/5 of hot).
• Latency of minutes to several hours for retrieval (varies by provider and tier).
• Per-GB retrieval fee on every restore.
• 90-day minimum storage duration typically.

Used for long-term backups, compliance archives, raw data preserved for occasional reprocessing.

What archive / deep archive gets you

• Lowest possible storage cost (often 1/20 of hot).
• Retrieval takes hours (standard) or up to 12+ hours (bulk).
• Substantial per-GB retrieval fees.
• 180-day or longer minimum storage duration.

Used for regulatory archives that legal requires you to keep but no one expects to ever read. Tape-replacement workloads. Cold-side disaster-recovery copies.

The minimum-storage-duration trap

This is the biggest gotcha. Cold and archive tiers charge for a minimum period regardless of actual storage:

• Store 1 TB in archive for 1 day, delete. Bill: full 180 days of storage (minimum duration).
• Migrate 50 TB to cold for a project, delete after 1 week. Bill: 90 days of storage for all 50 TB.

Tiering data that turns over frequently — even at apparent cost savings — backfires. The minimum-duration penalty wipes out the savings.

The retrieval-fee trap

Archive retrieval fees can dwarf storage costs:

Numbers are approximations. The point: the more often you retrieve, the smaller the archive savings, until eventually hot is cheaper.

Tiering policies and lifecycle rules

Cloud providers automate tier transitions via lifecycle rules:

• Object created in hot.
• After 30 days of no access, transition to warm.
• After 90 more days, transition to cold.
• After 365 more days, transition to archive.
• After 7 years total, delete.

Some services (AWS S3 Intelligent-Tiering, GCS Autoclass) monitor actual access patterns and tier dynamically without fixed schedules. Useful when access patterns are unpredictable; comes with its own per-object monitoring fee.

On-premises tiering

The same patterns apply to on-prem storage:

• NVMe SSDs for hot data — application working set, databases.
• SATA SSDs for warm data — backup catalogs, recent archives.
• HDDs for cold data — long-term file storage, media archives.
• Tape for archive — true long-term retention with very low per-GB cost.

Modern storage platforms (Synology Hyper Backup, TrueNAS, Synology Hybrid Share) automate tiered storage with similar lifecycle rules.

The hybrid pattern

Many production deployments use hybrid local + cloud tiering:

1. Hot data on local SSDs.
2. Warm data on local HDDs.
3. Cold backup copies in cloud cold/archive tiers.
4. Lifecycle rules transition between cloud tiers automatically.

This combines on-prem performance for active data with cloud economics for long-term retention. The catch is the egress-cost moment: bringing cloud-archived data back can produce surprising bills. See cloud egress costs.

Frequently Asked Questions

What is the difference between hot and cold storage?

Hot storage is optimized for frequent access — high IOPS, low latency, higher per-GB cost. Cold storage is optimized for infrequent access — lower per-GB cost but longer retrieval times and per-retrieval fees. Between them sit warm tiers (occasional access) and archive tiers (rare retrieval, hours of latency). The economic choice depends on how often the data will actually be read.

What is glacier-class storage?

The coldest archive tier from major cloud providers — AWS Glacier Deep Archive, Azure Archive, GCS Archive. Storage cost is roughly 1/10 of standard cloud storage. Retrieval takes hours; per-retrieval fees and minimum storage durations are substantial. Used for compliance archives, long-term backups, and any data that may never be read but must be retained.

When does cold storage save money?

When the data is retrieved rarely enough that storage savings exceed retrieval fees. Rough rule: if you'd retrieve less than once or twice per year, cold or archive tiers win. If you'd retrieve monthly, warm tiers are usually best. Hot is only correct if access is regular. Run the math: monthly storage cost × months held vs retrieval cost × expected retrievals.

What is minimum storage duration?

Some cold and archive tiers charge for a minimum storage period even if you delete data sooner. AWS Glacier Deep Archive charges 180 days minimum; AWS S3 Glacier Instant Retrieval has 90 days. Deleting after 30 days still incurs the full minimum-duration storage fee. Designed to discourage churn that doesn't fit the tier's economic model.

How do tiering policies work?

Cloud storage services let you define lifecycle rules that automatically transition objects between tiers based on age or access patterns. A typical policy: hot for 30 days, then warm for 90 days, then cold for 1 year, then delete or archive permanently. Some services (AWS S3 Intelligent-Tiering, GCS Autoclass) monitor access patterns and tier automatically based on observed usage rather than fixed schedules.