Questions

What’s wrong with your server or storage?

Find the cause, work through the checks in order, and compare compatible replacements only where hardware is actually the fix. Each page opens with the answer, maps the symptom you see to a diagnostic path with real commands, and says per cause whether buying anything helps.

24 pages · every technical statement cites a vendor or project document read on a stated date · section updated 2026-10-14

Enterprise SSD or NVMe

Wear, endurance, detection and firmware problems with U.2, U.3, E1.S and M.2 drives in servers.

Which SSDs should I buy for Ceph OSDs and BlueStore DB/WAL, and how big should they be?
Buy enterprise-class SSDs whose maker documents power loss protection, rated 1 DWPD or more (or the TBW equivalent): Ceph's Squid hardware page says many OSD deployments need no more than that and calls 3 DWPD mixed-use drives often overkill.
Do I need enterprise SSDs for Proxmox, or will consumer drives do?
Proxmox's current admin guide says 'Using consumer SSDs is discouraged' and recommends SSDs with power-loss protection (PLP) for good performance, while its 2020 ZFS benchmark paper answers the consumer and pro-sumer question with an unqualified no.
Enterprise SSD Percentage Used hit 100%: is it failing, and do you replace it?
Not by itself.
How much power does an enterprise U.2 NVMe SSD draw, and can I make it idle lower?
Take the figure from the maker's sheet or your server vendor's thermal table, not a forum: Kioxia lists its 2.5-inch CD8-R and CD8-V at 11-20 W typical active and 5 W typical Ready, and Dell's R760 table for rear NVMe drives lists figures from 14 W (Samsung PM9A3 7.68TB) to 25 W (Kioxia CM7 30.72TB).
Why does my NVMe SSD disappear under load until I power-cycle?
'controller is down; will reset: CSTS=0xffffffff' means the Linux nvme driver read the drive's status register back as all ones: the host could not reach the device at all, and when the reset that follows fails, the driver gives up and only a power cycle brings the drive back.
Proxmox shows high SSD wearout: is the drive dying, and what is writing to it?
On NVMe, the Wearout column is the drive's own SMART Percentage Used, a vendor estimate of life used that does not by itself mean the drive is failing, even at 100; on SATA it is 100 minus a vendor-normalised wear attribute.
Sabrent Rocket Enterprise or a consumer Sabrent Rocket in a server: what actually differs?
Yes, on paper they are different drives: Sabrent's manual describes the Rocket Enterprise as a 2.5-inch, 15mm U.2/U.3 PCIe 4.0 x4 drive rated 1 drive write per day, with power loss protection and DDR4 DRAM, while the consumer Rocket 4 Plus is an M.2 2280 drive rated 1,400TBW at 2TB and 2,800TBW at 4TB, and we found no Sabrent document stating power loss protection for the Rocket 4 Plus or the Rocket 5.
Is a QLC Solidigm D5-P5336 a sensible drive for a write-heavy workload?
It depends on the size and volume of the writes: Solidigm positions the QLC D5-P5336 for read-intensive work and calls TLC the fit for mixed and write-heavy work.
Why does my U.2 drive not show up in a U.3 bay?
If the bay is U.3-only, wired to a tri-mode controller over one shared link, the U.2 drive is the cause: SFF-TA-1001, the U.3 specification, does not require a U.3 slot to accept a U.2 drive, and its interoperation table gives "No link" for every U.2 drive in every U.3 slot width.
Why is my U.2 NVMe SSD not detected in my server?
A U.2 drive the server cannot see has lost one link in a chain: an NVMe-wired bay or a correctly split slot, a cable that carries the NVMe pinout, the PCIe link, then the Linux nvme driver.
Bought a used enterprise SSD? How to check its health and decide whether to keep it
Read the drive's own records before you put data on it: confirm the model, serial and firmware match the listing, then check Critical Warning, Available Spare against its threshold, media errors and the self-test log, and subtract the writes it has already logged from the rated TBW or DWPD for its exact model.

SAS/SATA drives and controllers

Drives that are not detected, sector-size and 4Kn problems, HBAs, RAID cards, cables and backplanes.

Why won't my server see or boot a 4Kn SAS drive?
A 4Kn drive presents 4096-byte logical sectors with no 512-byte emulation, Western Digital warns that older systems may not be compatible, and Dell, Intel and Supermicro each say 4K drives boot only in UEFI mode.
How do I fix SAS drives stuck at 520-byte sectors?
A 520-byte sector is a 512-byte sector with 8 bytes of checksum or protection information appended; NetApp's own patent describes 520-byte sectors used this way, and Linux rejects 520 as a block size.
HBA or RAID controller for TrueNAS: can you keep the card you have?
TrueNAS and OpenZFS both want each physical disk handed to ZFS directly: TrueNAS says not to use a RAID card's RAID facility and to use its HBA mode if it has one, and OpenZFS says hardware RAID controllers should not be used with ZFS.
Are refurbished enterprise hard drives worth it once the warranty and the failure risk are counted?
They can be, but only your own inputs settle it: the price gap against the same drive new, the warranty that actually attaches to the serial (six months from Seagate on its factory-recertified drives, one year from WD on its recertified store, otherwise whatever the seller promises) and a failure rate you choose, run through a cost per usable TB-year for each offer.
SAS expander or a 24-port HBA: which does a 24-bay chassis need?
If the backplane already has an expander and the drives are hard disks, an 8-port HBA's two x4 connectors carry them: by our SAS-link arithmetic from Western Digital's and Broadcom's figures, 24 Ultrastar HC550 drives at full sustained rate total 6,456 MB/s, more than one 12Gb/s x4 uplink (4,800 MB/s each way) but less than two, so cable both (we read no 8-port card's PCIe rating).
Why is my SAS hard drive not detected by the HBA?
When a SAS drive does not appear anywhere, check four things before you blame the drive: the port (Seagate states SAS drives cannot be plugged into SATA controllers), the cable (multi-lane end at the controller, SFF-8482 end at the drive, and the right direction into a backplane), the power-disable pin P3 (a supply holding 3.3V on it keeps a drive with that feature in reset, so it never spins up) and the controller mode (many SAS RAID controllers want a drive assigned to an array before the operating system sees it).

Server compatibility and memory

Third-party parts in Dell, HPE and Supermicro servers, ECC RAM that is not recognised, population rules.

Why do the fans ramp up and iDRAC show nothing after I fit a non-Dell NVMe SSD in my PowerEdge?
If the drive works in the OS, it is most likely iDRAC handling a device it does not recognise, not a faulty drive: Dell says iDRAC may spin the fans up for an unknown device and that it cannot support third-party devices, and in the threads below the drive works in the OS while iDRAC shows no health or firmware.
Why does my multi-drive NVMe card show only one drive, or none?
A passive multi-drive card gives each drive its own x4 share of the slot's lanes, so the slot has to be split in the firmware: Supermicro documents that a two-drive card in a slot left at x8 or x16 shows one drive, at x2x2x2x2 shows none, and shows both only at x4x4 (x4x4x4x4 in an x16 slot).
RDIMM or LRDIMM: which should I buy to add memory to my server?
Buy the class that is already fitted: Dell's R740 manual, Lenovo's SR650 guide and Intel's S2600WF rules all forbid mixing RDIMMs with LRDIMMs, so changing class means replacing every module rather than adding to them.
Why is my server not recognizing its new ECC RAM?
Dell's PowerEdge R740 manual lists the three ways a configuration that breaks its memory guidelines shows up: the system 'might not boot, stop responding during memory configuration, or operate with reduced memory.' The rules that can be broken are module class (RDIMM mixed with LRDIMM), generation, rank, width or capacity limits, and population order, and each platform sets its own, so check your server's own manual rather than a general rule.

Slow storage, networking and backup

Ceph and ZFS performance, 10GbE that did not get faster, LTO drives and media.

Why did my 10GbE upgrade not make file copies faster?
A 10GbE link only raises copy speed when every other stage of the copy keeps up with it: the drives behind the share, the SMB session, the client's own port and the PCIe slot each card sits in.
Is my Ceph cluster slow because of consumer SSDs, and how do I prove it?
It is a documented cause: Ceph's BlueStore opens its drives with O_DIRECT and calls fsync() frequently, and Ceph's documentation says enterprise-class SSDs are best for Ceph because they almost always have power loss protection and do not suffer the cliffing client (desktop) models may show once a limited cache fills.
How big should the SSDs in a ZFS special vdev be, and should you add one at all?
We found no OpenZFS, Proxmox or TrueNAS document that gives a sizing rule for a special vdev, so size it from your own pool: zdb's block statistics show how much metadata the pool holds and how much data sits in small blocks, and zpool list -v shows how full an existing special vdev is.

How these pages are built

Each page is researched from vendor and project documentation and from discussions we opened and read, written by a storage engineer’s standard, then reviewed against its sources before it is published. Forum threads are evidence that a problem occurs; specifications and compatibility statements come only from primary documents, and every source carries the date we read it.

Where a page shows products, they are the live Amazon US listings in our catalogue that meet every criterion the diagnosis sets, with condition, price and stock shown as they are now. A listing that does not state a required field is left out rather than assumed. When the fix is a setting, a cable or a format, the page says so and shows nothing to buy.