Correctable Memory Errors on a Latency-Sensitive SQL Server
An R750 runs a latency-sensitive SQL Server instance. It has two Xeon Gold 6338 CPUs and 16 x 64GB DDR4 RDIMMs, eight per CPU, in slots A1-A8 and B1-B8. iDRAC alerts show 'Correctable memory error rate exceeded for DIMM_B4' 37 times in 24 hours. The SEL also contains several correctable ECC entries for the same slot, but no uncorrectable errors. The operating system event log has no machine-check exceptions. The server is under a 99.99% SLA, and the next maintenance window is 11 days away. A junior administrator proposes pulling DIMM_B4 now and running with 15 DIMMs until the replacement arrives because 'the server has spare memory.' The customer also asks whether enabling rank sparing in BIOS will make the errors disappear.
1. Assess the risk of waiting 11 days versus replacing DIMM_B4 now. What do the correctable errors indicate, and what data would you collect before touching hardware?
Read the worked answer
Correctable ECC errors are corrected, but they are not automatically harmless. A rising rate on a single DIMM usually indicates a degrading DIMM, although the same signature can come from the slot, the memory controller, or a firmware bug. The risk of waiting 11 days is that the rate may increase or cross into uncorrectable errors, which can crash the instance or corrupt data. For a 99.99% SLA and a latency-sensitive database, even corrected errors can add jitter. Do not wait passively: collect a SupportAssist bundle, the SEL, iDRAC memory health and error graphs, BIOS and iDRAC versions, and the DIMM part and slot inventory. Check whether the rate is stable or accelerating. If it is stable, no uncorrectable errors are present, and a replacement DIMM is on site, schedule replacement at the earliest window. If the rate rises, a uncorrectable error appears, or the same slot produces new symptoms, escalate to immediate replacement. Run the ePSA memory test in a maintenance window, not during production.
2. Evaluate the proposal to run temporarily with 15 DIMMs. What are the performance, NUMA, and reliability consequences?
Read the worked answer
Running with 15 DIMMs is usually a bad trade. Removing B4 leaves CPU2 with seven DIMMs while CPU1 has eight, which breaks the balanced population and can disable or change memory interleaving. The host may boot, but CPU2 loses a channel's worth of bandwidth and the NUMA topology becomes uneven; SQL Server may see asymmetric nodes and suffer plan or performance changes. Capacity drops by 64GB, and the system loses redundancy. Some population rules may flag the configuration as unsupported or downclock the remaining DIMMs. Rank sparing is not a fix: it reserves capacity to absorb a failing rank and can mask the symptom, but it does not repair the failing DIMM and it reduces usable memory. The right action is to replace B4 with an identical Dell-qualified DIMM at the maintenance window. If a temporary replacement is unavoidable, use an identical part and confirm the population is supported; do not simply leave a hole.
3. Describe the replacement procedure and the follow-up checks that distinguish a bad DIMM from a bad slot or CPU.
Read the worked answer
Schedule a maintenance window, back up, and shut down gracefully. Power off, disconnect power, wear ESD protection, and locate DIMM_B4 using the system-board silkscreen and the iDRAC inventory. Replace it with an identical Dell-qualified DIMM, seat it firmly, and power on. Check the iDRAC memory inventory for capacity, speed, and slot status; run the ePSA memory test; clear the SEL and monitor for 24 to 48 hours. If the correctable errors disappear, the DIMM was the cause. If errors persist on the new DIMM in B4, move that DIMM to a known-good slot: if the errors follow the DIMM, replace it again; if they stay with B4, suspect the slot, CPU2, or the system board and engage Dell support with the SupportAssist collection. Also verify BIOS and iDRAC are at the recommended levels, because memory training and error-reporting fixes can change the behavior.