What PLP actually does
Most SSDs hold incoming writes in a volatile buffer before committing them to flash. They also keep their logical-to-physical mapping tables in volatile memory. A clean shutdown gives the drive time to save all of that. A sudden power cut doesn't.
Enterprise drives add onboard capacitors. These store enough energy for the controller to finish flushing buffers and updating the mapping table. Sherback says most consumer drives omit them mainly for cost and board space.
The important caveat is that PLP covers one specific failure window. It doesn't protect against bad firmware, a dead NAND die, or a deleted file. It is not a backup.
The sync-write angle
A synchronous write is software asking for a guarantee that data is on stable storage before it carries on. Databases do this constantly, and NFS and ZFS do it too. OpenZFS documentation puts it in terms of fsync() and O_SYNC. Those calls must reach stable storage before returning, and the ZFS intent log exists to satisfy that.
OpenZFS also says a dedicated log device (SLOG) is not a write cache. Asynchronous writes never touch it. If your workload has no synchronous writes, adding a SLOG changes nothing. The same documentation says a log device should be low-latency and have power-loss protection. It warns that consumer SSDs without PLP will happily report data as stable that a power cut then loses. The typical beneficiaries it names are NFS servers, databases and VM hosts with sync-heavy guests.
That is the performance side of Sherback's argument. A drive without PLP can only commit to flash so fast, and sync-heavy loads expose that. A PLP drive can treat its protected buffer as safe and acknowledge writes sooner. A drive without it must wait for the flash itself to be written.
What Proxmox recommends
Sherback points to Proxmox's system requirements, and the documentation backs him up. The Proxmox VE wiki recommends SSDs with Power-Loss-Protection (PLP) for good performance, and says using consumer SSDs is discouraged. A Proxmox staff reply on its forum goes further. It highly recommends PLP SSDs for ZFS and Ceph, and warns that some WD Red disks marketed as Enterprise lack PLP, so buyers should double-check.
That warning matters for second-hand shoppers. The word "enterprise" on a listing doesn't guarantee the feature.
Third-party guides repeat the advice. Cherry Servers, for instance, says enterprise SSDs with PLP can deliver tens of times more sync write IOPS than consumer drives without it. That is a vendor blog's claim, not a measured benchmark from this story, so treat the multiple as workload-dependent.
A nuance for ZFS users
Sherback says OpenZFS guidance wants power failure protection covering both flushed data and the drive's own metadata. That is correct, but the guidance is more nuanced than "you need capacitors".
- ZFS only needs the drive to protect flushed data and its own metadata through a power cut.
- Some manufacturers claim firmware-based protection achieves this without capacitors.
- Hardware PLP also tries to protect unflushed data. OpenZFS says this gives ZFS no extra benefit, though it does no harm.
- The documentation keeps non-exhaustive lists of NVMe and SAS drives with power failure protection.
So a capacitor drive is the safe default for ZFS, particularly as a SLOG. It isn't the only conceivable design that satisfies ZFS. The model and its documentation matter more than the label.
How to vet a used drive
Sherback says a seller who won't share a health readout is probably not worth buying from. That is sound. Here is a practical way to apply it.
- Confirm the exact model has PLP. Check the manufacturer's datasheet for the part number. Don't rely on the product family or the word "enterprise".
- Ask for a SMART or health report. Look at wear, power-on hours and error history.
- For NVMe drives, check the Critical Warning field. In the NVMe definitions, no wait, see below
Correction to step 3: the NVMe Critical Warning flag for the volatile-memory backup device is documented in the Linux nvme definitions. Per that documentation, if the bit is set then the volatile memory backup device has failed. The field is only valid if the controller has a volatile memory backup solution. A clean reading is good evidence, but it isn't a full health guarantee. Drives don't all report capacitor state the same way.
- For SATA drives, check the model documentation. Sherback says Intel's SATA data center drives log a capacitor self-test result. I couldn't verify a universal attribute name or procedure for this, so check your specific model's documentation rather than assuming every SATA enterprise drive exposes it.
- Check the seller's return policy. Used drives rarely carry a warranty.
The baggage that comes with used drives
Sherback is upfront about the downsides:
- Warranty: there is likely none, and you are trusting the seller.
- Interfaces: SAS drives need an HBA. U.2 drives need adapters and real airflow.
- Power: enterprise drives tend to draw more at idle than consumer ones.
- Firmware: drives pulled from OEM servers can carry vendor firmware that is hard to update.
The airflow point deserves emphasis. OpenZFS gives one model-specific example of an enterprise NVMe drive whose manual specifies airflow requirements. Without enough airflow, that drive overheats and throttles severely. In the worst case it suffers a "degraded reliability" event that makes a namespace unusable until a secure erase. That is one model, not a rule for all enterprise SSDs. It does show why you should read the thermal requirements before buying.
UPS versus PLP
Sherback separates the two protections well. A UPS covers the wall outlet but does nothing if the PSU itself dies. PLP covers the drive's own shutdown handling but doesn't keep the host running. OpenZFS also recommends stable power through UPS units or line conditioners, because brownouts are a risk of their own. They complement each other.
Where the argument is strongest, and where it isn't
Strongest: a Proxmox or ZFS host that runs unattended, handles sync-heavy VM, database or NFS writes, and holds data you care about.
Weaker: a gaming PC or ordinary desktop. Sherback himself would still buy a new consumer drive there, and that is reasonable. Consumer drives aren't categorically unsafe, and they come with a warranty. Mostly asynchronous desktop writes don't see the sync-write benefit.
Worth keeping in mind:
- Used drives have unknown wear, so a clean report is a screening aid rather than proof.
- A warranty on a new consumer drive replaces the drive, not the data on it. That is why backups still matter.
- Buy for the exact model's documented features, not the category label.
The bottom line
The takeaway is narrower than the headline but still useful. If you are building a home server with VMs, databases or ZFS, a PLP drive is worth paying attention to, and used enterprise stock may be the cheapest way to get one. Verify PLP for the specific model, read the health data, plan for cooling and connectors, and keep real backups. A UPS is a sensible addition too, as Sherback concedes about his own setup.