Storage and transfer: account for volume over time and path

A pool containing two hundred GiB at period end may have held that amount for one day or the entire period. Integrate retained volume with its actual time intervals, including the replicas, snapshots, and versions inside the chosen billing boundary. Logical dataset size can differ from allocated storage because of replication, metadata, encoding, or minimum object sizes. State which quantity the hypothetical rate prices. Preserve the deletion and retention policy as well: a request to delete an object need not establish when its retained versions cease to occupy the accounted resource.

Choose units before calculating. A GiB is 2³⁰ bytes, while a decimal GB is 10⁹ bytes; the label alone must not be silently changed between measurement and pricing. Similarly, an export can represent byte-seconds even when a price is displayed per a larger capacity-time unit. Convert both capacity and duration consistently. This lesson uses GiB and days throughout its storage integral, with an explicitly thirty-day teaching month. That convention is an arithmetic fixture, not a claim about calendar billing, free tiers, minimum commitments, or real invoice rounding.

List request classes separately when they have different units: object reads, writes, listing operations, and retrieval operations need not carry the same policy. A byte total cannot reconstruct their counts. A compressed transfer can have different wire bytes from the decoded object size, and retries can increase the transmitted quantity. Record route, source, destination, and direction rather than label all movement as egress. Internal replication, cross-zone traffic, external delivery, and local reads may have different rules or no separate charge under a particular documented service.

Avoid counting a packaged charge again through its underlying operations. If a fictional storage fee already includes replication, adding another replica fee without a separate boundary double-counts. If a transfer crosses multiple charged segments, retain each eligible segment and its own quantity instead of assuming one end-to-end price. Reconcile the measured quantities with the pricing-unit fields and charge types of the chosen ledger. The useful inventory is a reproducible route and operation map, not an extrapolation from an interface’s advertised link bandwidth.

A smaller retained checkpoint set can lower capacity-time while increasing recovery recomputation or repeated transfer. A remote cache can reduce model execution but add network movement, lookup operations, and another failure dependency. Compare the complete accepted workload and recovery requirement under each policy. State which data must remain available and for how long; deleting required evidence is a changed service, not an equivalent storage optimization. Include failed transfers and retry routes when those resources fall inside the chosen accounting boundary.

The function lab illustrates normalized volume-time. Reset, explicitly set a=0.75 and x=2, predict the area, then Play or Step. Repeat from a fresh Reset with a=1.5 and x=2 explicitly set; the areas under y=ax² are two and four. These are normalized area units rather than a real storage bill. The instrument has no objects, deletion policy, operations, transfer route, or billing month. Its purpose is to show why the whole retention curve matters rather than just its ending volume.