Hot storage tier
A hot storage tier prioritizes the access demands of active data through its performance, capacity and placement policy.
Frequent access and low latency are separate requirements. A rarely read object may still need an immediate response. Classify data using the application's access contract rather than age alone. [4]
The hot system in the f4 study used HDDs, as did the warm system. Different layouts and redundancy policies supported different demand levels; hot did not mean flash. Cached requests also changed the load reaching storage. [2]
Relations
Advantages
- Compact in-memory metadata can reduce extra disk operations during blob reads. [2]
Limitations
- Capacity-efficient placement must still sustain demand bursts and recovery traffic. [2]
- An infrequent-access class may retain fast reads but add retrieval charges, so frequency and latency cannot be treated as synonyms. [4]
Citations
- f4: Facebook's Warm BLOB Storage System, Muralidhar et al., USENIX OSDI 2014
- Amazon S3 User Guide: Understanding and managing S3 storage classes
Further reading
- Hot storage tier
- NVM Express NVM Command Set Specification, Revision 1.3 (31 July 2026), NVM Express, Inc.
Sources & caveats
- The f4 paper describes Facebook's historical immutable-blob architecture, not a universal hot-tier threshold or a current hardware benchmark.
- The previous claim that the f4 paper contains no hot-tier latency figure was removed: section 6.3 reports measured local-read latency.