Cold, Warm, and Hot Storage Tiers

How Kronos places data across access tiers so active pulses are fast to reach while the full history stays affordable to keep.

Not all data needs to be fast

Petabyte histories cannot all sit on the fastest media, and they do not need to. Kronos tiers storage by access temperature: hot data close to compute for immediate use, warm data on capable but slower media, cold data in deep archive for long-term retention. Placement matches how likely and how urgently each record will be read.

The three temperatures

Hot storage holds the pulses and datasets an active campaign is training or refining against, on fast NVMe near the GPUs. Warm storage holds recent and frequently referenced history on capacity media with good bandwidth. Cold storage holds the long tail, older pulses and completed campaigns, retained for audit and occasional replay but rarely touched.

Movement between tiers

Data migrates by policy and by demand. A pulse cools from hot to warm to cold as it stops being referenced; a cold pulse pulled by a refinement study is staged back to hot for the duration. This staging is handled below the workload, so a study requests a pulse and the tiering system makes it fast without the user managing media.

Tiering is what keeps petabyte-scale retention sustainable while keeping active work fast. If every pulse demanded hot storage, the archive would be impractical; if every read hit cold storage, campaigns would stall. Tiering resolves the tension by matching placement to actual access patterns.

The tiers serve both machines and both data kinds, hardware pulses and simulation campaigns, under one policy. Immutable and hashed at every tier, records keep their integrity as they move, so a cold-archived certified neutronics result is exactly as trustworthy as the day it was written when it is eventually recalled.