Re: Best way to completely test (non-SMART) drive in NetBSD ?

[Greg Troxel <gdt%lexort.com@localhost>]

What I do is write the entire disk and read it back.  I wrote a program,
which I should perhaps package, that writes mostly zeros but each block
has the block number.  on read it looks for that.   however, I have
essentially never, maybe never in doing these tests, had issues with a
read returning a non-error status and the wrong data.

random writes can be very slow (because urandom has to generate).


I would thus recommend:

  dd if=/dev/zero of=/dev/rsd4c bs=32k
  # actually try a few bs= all 2^N and find the smallest one such that
  # doubling doesn't really help.  I tend to use 1m, but that's on
  # computers from >=2010 with at least 1G of system RAM.

  dd if=/dev/rsd4c of=/dev/null bs=32k

of course be super careful not to zero the wrong device.   That's a
major hesitation for me in publishing code.

Then maybe power it down for a few days, start it up, and read again.

Another suggestion is to read first, before you write.  When a block
goes bad, the controller detects it and replaces it with a spare, but
you get a forced error on the read until you write it.  If this happens
for a few blocks once in a great while it's sort of ok but once more
than rare it's a sign of trouble and the disk is not going to be ok.


I have a script that does reads via dd of all my on-line disks across
all systems, and I run it every 2 months.  Mostly they are ok; there is
one old disk that needs some bad block fixups.  I'd replace that disk
normally but for reasons you are unlikey to share it's hard.