Which SCSI interface/adapter are you running?
- LSI PCI-X SCSI U320 HBA (brand new)
- Amphenol LVD 2m cable (brand new)
- Amphenol U320 LVD terminator (brand new)
- Rax 3xSCA disks with temperature control (brand new)
Why do you ask this? Assuming the kernel driver is working fine(2), if { HBA, cable, terminator } is the problem, then you should see increasing the "non medium error" value.
It didn't happen with the new setup, hence the LVD-setup is fine; indeed, there are Fujitsu 10K rpm and Seagate 15K rpm disks that have perfectly passed all the "badblocks" tests (8 hours burn-in)
Good point, however
I will add a function to check the "non-medium error" value, in order to stop the test program if the value is seen to increase due to a bad physical SCSI configuration rather than due to a physical problem(1) with the disk under testing.
(1) physical problems that I cannot test directly appear on the SCSI interface (hence to my testing program) like a communication/service disruption
- worn out bearings -> read/write delayed correction or I/O abort
- worn out read/write heads -> read/write delayed corrections or I/O abort
- worn out brush less motor -> read/write delayed corrections or I/O abort
- worn out SCA connector -> wrong "tag phase reported" by the Linux kernel, disk not recognized, channel too noisy with too many retries (you have exactly this symptom if you use a bad cable, or a bad SCSI terminator)
- worn out electronic board with semi fried chips -> read/write delayed correction or I/O abort
edit:
Yup, I also have to write a function to monitor what the kernel complains for. Not yet done.
(2) there are certain SCSI devices { CDROM, DVDRAM, MO, CD-Jbox, Tape{DDS, LDO, ...}, Scanner } that have SCSI-quirks, but I have never observed anything similar with SCSI disks.
When a quirk arises, you see the kernel complain about phases and tags, with a lot of verbosity.
If you don't see it, it's mostly 99.97% ok.