Warm storage tier
A warm storage tier holds less frequently accessed data while retaining the access behavior required by its workload.
Warm is a relative workload category, not a fixed disk type or IOPS-per-terabyte limit. The f4 design used online, erasure-coded HDD storage for immutable blobs with reduced request demand. [2]
Its proposed analytical cutoffs differed from the deployed policy. Section 6.2, footnote 4, reports approximately three months before migration for simplicity. Age is a workload-specific proxy, not a universal definition. [2]
A cold tier can also be online. Check the actual retrieval contract rather than assuming every colder tier imposes an hours-long wait. [4]
Capacity accounting: the retained volume of deleted-but-unreclaimed data matters alongside the redundancy factor. A worked audit of f4's printed equation separates exact arithmetic, early rounding and the unknown deployment mixture. [5]
Relations
Advantages
- Erasure coding can reduce stored redundancy overhead compared with full replication. [3]
Limitations
- Reconstruction after unavailable fragments adds work and can affect read latency. [3]
- Mutable data and renewed demand require a separate placement policy; historical immutable-blob cutoffs are not transferable guarantees. [2]
Citations
- Muralidhar et al., f4: warm BLOB storage system, OSDI 2014
- Huang et al., Erasure Coding in a cloud storage service (Local Reconstruction Codes), USENIX ATC 2012
- Microsoft Learn: Access tiers for blob data, online tiers, archive and billing
- Hesela: f4 capacity arithmetic audit, exact fractions and reproducible code
Further reading
Sources & caveats
- The f4 and Azure LRC papers concern specific large-scale systems, not standard temperature thresholds.
- Azure Blob Storage's online Cold tier is a counterexample to treating cold as a synonym for offline archive.
- The calculation substitutes exactly 65 PB for illustration; the source reports a corpus larger than that. It does not reconstruct the reported 53-PB saving.