On Thu, Nov 04, 2021 at 10:43:23AM -0700, Christoph Hellwig wrote: > Well, the answer for other interfaces (at least at the gold plated > cost option) is so strong internal CRCs that user visible bits clobbered > by cosmic rays don't realisticly happen. But it is a problem with the > cheaper ones, and at least SCSI and NVMe offer the error list through > the Get LBA status command (and I bet ATA too, but I haven't looked into > that). Oddly enough there has never been much interested from the > fs community for those. "don't realistically happen" is different when you're talking about "doesn't happen within the warranty period of my laptop's SSD" and "doesn't happen on my fleet of 10k servers before they're taken out of service". There's also a big difference in speeds between an NVMe drive (7GB/s) and a memory device (20-50GB/s). The UBER being talked about when I was still at Intel was similar to / slightly better than DRAM, but that's still several failures per year across an entire data centre that's using pmem flat-out.
