What to Check When Buying Used Enterprise Drives

On a hard drive the figure is power on hours, and it is recorded by the drive itself rather than by anyone selling it. It counts the time the drive has been running, which in a data center normally means continuously since the day it was racked.

There is no universal threshold, and any supplier quoting one as a rule is inventing it. Enterprise drives are specified for continuous operation, so high hours on their own are not a fault. What hours tell you is how much of a design life has been consumed, which helps you decide what to pay rather than whether to buy.

The number becomes decisive read next to the failure indicators. A high hour drive with a completely clean sector count has demonstrated durability. A low hour drive that has already started reallocating has demonstrated the opposite, and the second one is the one to worry about even though its headline number looks better.

For flash, hours are the wrong measure entirely. That is the next section.

Three SMART fields answer this on a hard drive, and they are worth knowing by name because they are the difference between a good pallet and an expensive one.

Reallocated sector count is the number of sectors the drive has found bad and remapped to spare capacity. A drive with a rising count is failing quietly and covering its own tracks. Any non zero value deserves attention, and a large one deserves rejection.

Current pending sector count is worse, because it counts sectors the drive has flagged as suspect but has not yet been able to remap. Pending sectors mean unresolved read problems right now.

UDMA CRC error count is the odd one out and gets misread constantly. It counts data errors on the cable or backplane rather than on the platters, so it usually points at a bad cable, a dirty connector or a marginal backplane slot rather than at a dying drive. A lot with widespread CRC errors is telling you something about the system the drives came out of, which is worth knowing.

Read all three together. Any one of them alone can mislead.

A solid state drive does not wear out by running. It wears out by being written to, because each flash cell tolerates a finite number of program and erase cycles. That single difference invalidates most of the intuition people bring from spinning disk.

The consequence is that power on hours tell you almost nothing about a used SSD. A drive that has been powered on for years serving mostly reads can be barely worn. A drive with a fraction of those hours in a write heavy logging or database role can be close to the end of its rated endurance. The one with the lower hours is the worse buy, and nothing on the outside of either drive shows it.

The figure that does answer the question is the percentage of rated endurance consumed, which most enterprise drives report directly, often alongside a total of host writes. That is the flash equivalent of power on hours and it is the number to ask for.

If a seller offers hours on flash and nothing else, they are either not reading the drives or not reading the right field.

Enterprise flash is built in classes for different write loads, normally described as read intensive, mixed use and write intensive in ascending order of endurance. The class is designed into the drive through the amount of spare capacity held back and the type of flash used, and it cannot be changed afterwards.

The units used to express it are drive writes per day over the warranted life, and total terabytes written. Both appear on the manufacturer's datasheet for the exact part number, which is why flash is quoted against the part number rather than the capacity. Two drives that both read the same capacity on the sticker can sit in different endurance classes and be very different items.

This matters when you are buying used, because a read intensive drive bought cheaply and deployed into a write heavy role will consume its endurance far faster than the price suggested. The drive is not faulty and the seller has not misled you. The class was simply wrong for the job.

Decide what the drives will be doing before you decide what to pay, then match the class to it.

Data you cannot check is a claim rather than a specification, and the whole point of the first two questions is that both are answered by the drive itself rather than by the person selling it.

What to ask for is the underlying figures rather than a summary. A grade or a health percentage is somebody else's interpretation of the data. The raw fields let you apply your own thresholds, which will differ from theirs, because a rebuilder replacing drives in a rental fleet and an operator filling a nearline array have different tolerances for the same numbers.

GTR Supply supplies a SMART health report per drive on hard drive lots, and both the grade and the underlying test data go on the manifest, so you can work from the numbers instead of the adjective.

When you receive a lot, spot check it yourself rather than only reading the paperwork. Reading SMART data takes seconds per drive with standard tooling, and checking a sample against the manifest on arrival is the cheapest quality control available to you.

A lot can be healthy and still be the wrong purchase, and these three account for most of it.

Sector format. Drives ship in 512e and 4Kn variants, sometimes under the same model family, and mixing formats inside one array is the expensive mistake. Check the part number rather than the capacity, and check what your existing array is running before you buy anything to extend it.

Caddies and sleds. A drive in its original sled goes straight back into the chassis family it came from. A bare drive needs a tray bought separately, and trays for some platforms cost more than a used drive is worth. Decide which you are buying before the price is agreed, and if the drives are already out of the sleds, keep the screws with them.

Firmware. Some arrays and controllers are particular about drive firmware and will flag an otherwise perfect drive as unsupported. If the drives are going into a specific platform, say which one at quote stage rather than after delivery.

Mixed pallets are normal and perfectly buyable if you price them for what they are rather than for the best line on them.

Start by getting the manifest split by capacity and interface with a count against each, because a total drive count with no breakdown is not something anyone can price. Then value the pallet as its parts. Identify the lines you actually want, decide honestly what the remainder is worth to you, including the possibility that it is worth very little, and pay for the sum rather than the average.

Be realistic about the small legacy capacities. They take longer to place whatever their condition, and a pallet whose count is padded with them is worth less than its drive count implies.

Ask whether the pallet has been picked over. A lot nobody has been through generally beats a larger quantity that has already had the good parts removed, because the parts taken out first are always the ones carrying the value.