Cold archive tier
A cold archive tier is a storage class for rarely accessed data whose retention and retrieval requirements govern its design.
Cold and archive are not interchangeable access guarantees. One block-blob service keeps Cold online but requires Archive payloads to be rehydrated before reads. Some metadata remains available during archiving. [4]
Pergamum proposes cached-signature comparisons while disks are off; full scrubbing reads the disk blocks. Cached agreement cannot expose unread payload changes before a fresh media read. Section 5.3 excludes scrubbing and consistency checking from the evaluated prototype, so these mechanisms are not measured prototype capabilities. [3]
A minimum billable duration concerns charges, not a retention lock. Verify deletion protection, recovery copies and the restoration path separately from the storage tier. [4]
Object count and payload-size distribution also affect billing. Distinguish per-object size floors from separately billed metadata; neither specifies physical media allocation. [6]
Restoration also has a lifetime. S3 Glacier Flexible Retrieval and Deep Archive make a temporary readable copy without changing the object's storage class. S3 Intelligent-Tiering archive restoration instead moves the object to Frequent Access and does not accept a Days parameter. Check completion and expiry, not only request acceptance. [5]
Shared-resource limits also constrain archive access; a rack-wide count can miss a local conflict. Pelican's rules count both spinning and spinning-up disks. [2]
Relations
- related
- Hot, warm, and cold storage tiers, Magnetic tape storage, LTO (Linear Tape-Open), Object storage, Air-gapped backup, Warm storage tier, Hot storage tier, Data scrubbing, Storage resource domain, Minimum storage duration, Minimum billable object size, Algebraic signature, Lustre HSM, Lustre HSM release
Advantages
- Maintaining metadata in non-volatile memory can avoid some spin-ups for metadata operations. [3]
Limitations
- A powered-down disk may remain attached to a networked system, so power state alone does not establish an air gap. [3]
- Restoration delay and retrieval charges must be included in recovery planning. [4]
Citations
- Balakrishnan et al.: Pelican: A Building Block for Exascale Cold Data Storage, USENIX OSDI 2014 (research prototype paper)
- Storer et al., Pergamum, FAST 2008, sections 4.1-4.3.2 and 5.3
- Microsoft Learn: Access tiers for blob data, online tiers, archive and billing
- Amazon S3 User Guide: Working with archived objects, temporary copies and Intelligent-Tiering
- Amazon S3 pricing: object-size floors and archive metadata
Further reading
Sources & caveats
- Pergamum is the FAST 2008 disk-based research system. Sections 4.2 and 4.3.2 distinguish signature comparison from media scrubbing.
- The unobserved-payload limitation follows from the mechanism; no new hardware experiment is claimed.
- The Azure example concerns block-blob service semantics, not all archival media or a universal cold-access delay.
- The prototype limitation is explicit in section 5.3; mechanism coverage is separate from implementation evaluation.