Author: b t

A drive marked Failed is not necessarily a failed drive. Dropped links, SMART thresholds that hide degradation, simultaneous failures, slot numbering errors and backplane faults — five common misdiagnoses and the order to work through them.

Why unrecoverable read errors make RAID 5 risky on high-capacity drives; why throughput halves while every drive reports healthy; and the two highest-risk storage operations — controller replacement and running a dual-controller array on one controller.

A failed RAID cache battery (BBU, FBWC or supercapacitor) drops the controller from write-back to write-through and costs an order of magnitude on writes. Learn cycles, write-back that won’t re-enable after replacement, and what the battery actually protects.

One DIMM reporting correctable errors can downclock an entire channel; wrong DIMM population order costs capacity or bandwidth silently; and the hardest case is the uncorrectable error that reboots the machine with nothing in the OS log. Diagnosing all three from the BMC.

Three reasons an SSD slows down (SLC cache exhausted, low free space, endurance throttling); planning replacement from Percentage Used and total bytes written rather than age; and the batch firmware risk that makes a whole set fail together.

Optic vendor coding checks that keep a port dark; three reasons a 10G port negotiates at 1G; and intermittent packet loss that takes three days to trace to a patch cable. Using CRC and error counters to confirm a physical layer fault, and the cheapest-first replacement order.

A failed module in a redundant power supply is silent; unstable power looks exactly like failing drives; fans at full speed are usually not the fans’ fault; one stopped fan can run a CPU hot for weeks. Inspection and replacement intervals for both.