An SSD's Spare Capacity Has Its Own Denominator

Locate the reserve before interpreting the percentage

The host addresses logical blocks; the SSD controller maps them to physical NAND locations. Micron's address-mapping explanation describes this distinction. A host capacity figure therefore identifies the logical space presented by the device, rather than an inventory of every physical NAND location.

Kingston's March 2019 explanation describes manufacturer-set over-provisioning as capacity withheld for controller use and invisible to the host operating system. Its formula divides physical capacity minus user capacity by user capacity. Expressing that ratio as a percentage requires multiplying by 100.

Here, user capacity means the device's host-addressable capacity. It does not mean remaining filesystem free space or a tenant's allocation. The linked raw, protected and client-usable capacity guide addresses protection and allocation further along the storage path.

References: [M1][K1]

OP-243: 200 reserved units can mean 25% or 20%

OP-243 is an invented SSD design with exactly 1,000 equal capacity units of physical NAND. It exposes exactly 800 units to the host and permanently withholds the remaining 200 for controller use. These stipulated units describe no real product, advertised byte capacity or measured drive.

Using user capacity as the denominator, the reserve percentage is (1,000 - 800) / 800 x 100 = 25%. That says there are 25 reserved units for every 100 units exposed to the host.

Using physical capacity as the denominator, the reserved share is 200 / 1,000 x 100 = 20%. That says 20 of every 100 physical units are withheld from host capacity.

Both statements describe the same 200 units. A summary saying 25% relative to user capacity and another saying 20% of physical capacity do not establish different hardware. Comparing their percentages without their denominators would create a difference that is absent from the stipulated design.

The physical share can also be recovered from the user-based ratio: 0.25 / (1 + 0.25) = 0.20. This conversion is valid because both statements use the same physical, exposed and reserved populations. It cannot reconcile two suppliers that count different capacity or different kinds of reserve.

Keep the filesystem's 300 free units inside its 800 units

Now stipulate one simplified filesystem spanning OP-243's entire 800-unit host capacity. It reports 500 units used and 300 free. Formatting and metadata are excluded only for this example.

The device boundary remains 1,000 physical units, split into 800 host-addressable units and 200 permanently reserved units. Inside the 800-unit host population, the filesystem accounts for 500 used plus 300 free. Its free space has not moved outside that population.

Adding the 300 free units to the 200 hidden units and calling the result permanent controller reserve would mix two different boundaries. If the filesystem later fills those 300 units, the stipulated device still exposes 800 units and withholds 200. Its permanent reserve has not changed.

Permanent reserve here describes the amount withheld by the invented design. It does not stipulate that the same physical NAND locations stay empty or that every reserved unit is already erased. Neither the 200-unit reserve nor the 300-unit filesystem figure identifies the current number of erased physical blocks.

Unused host space can matter without becoming permanent reserve

The distinction does not make host free space irrelevant. KIOXIA's April 2023 managed-flash brief explains that host information about unneeded data can reduce the valid data that must be moved during reclamation. Whether that information reaches a particular SSD depends on the supported storage path; a filesystem's free-space total alone does not establish it.

The controller also needs physical working space while preserving still-needed data and reclaiming blocks. That explains why spare capacity can matter. It does not turn a capacity percentage into a count of blocks ready to accept writing at this moment.

The linked garbage-collection article explains valid-page relocation, block erasure and the difference between fresh and workload-conditioned performance. Those mechanisms connect reserve capacity to write behavior, but their actual cost must be established for the device and workload.

References: [K2]

Compare definitions before comparing the offered variants

For a real offer, first resolve whether its OP percentage uses host capacity or physical capacity as the denominator, and what the supplier includes in the numerator. A factory-withheld capacity figure and a description of unused host space do not identify the same reserve. A percentage without that definition leaves the comparison unresolved.

Keep the exact model, capacity variant and capacity basis attached to the offer. A rounded advertised capacity alone cannot establish installed NAND capacity; the physical amount in OP-243 is known only because the example explicitly supplies it. Do not infer an equivalent hidden population for an actual SSD from its label.

Once the definitions agree, assess the exposed capacity against the storage requirement. Then evaluate applicable performance and endurance evidence separately. Retain the selected variant and workload basis with any supplied ratings. Kingston's historical product comparisons do not provide a universal multiplier for another drive.

More hidden capacity by itself specifies neither a speed increase nor a lifetime extension, warranty entitlement or sanitization result. The endurance guide explains host-write ratings, while the workload-conditioned performance guide explains why starting state matters. The useful purchase is a supported configuration with sufficient exposed capacity and applicable evidence for the intended work; the largest OP percentage alone cannot identify it.

References: [K1]

Carry the selected hardware into the physical installation

Once the platform and storage owners have selected an actual supported server and storage package, its equipment and released connection plans can become inputs to Leviathan's rack-and-stack service. The linked service covers agreed equipment mounting, IT cabling, labeling and physical installation records.

Use the service's project inquiry to share a nonconfidential equipment summary, rack and connection plans, site constraints and required physical handoff. This article interprets capacity offers; the installation scope described here does not offer SSD reconfiguration, firmware tuning, data disposition or warranty service.

References: [L1]