On 9/7/22 08:09, Jan Stary wrote:
> > 1) On initial boot (with 7.1 release, on a usb stick) it more or less
> > immediately panicked into ddb when I tried to pipe dmesg into a file on
> > the usb stick. I took out the NVMe-card, and whether or not that was the
> > problem the machine anyhow behaved better long enough for me to get
> > network and do a fw_update.
>
> sure sounds like it could be a bad USB stick.
> Very common. For important things, I have learned to write zeros over
> the entire USB stick before expecting it to actually work. Nothing to
> do with the T5500.
I am puzzled: how exactly is a zero filled USB stick
less panicky than another USB stick?
My experience with floating gate storage (SSD, Flash) has been less
than stellar, and I'm a bit cynical about billions of microscopic
capacitors holding their charge reliably. Well, perhaps I'm TOO
cynical, but I've had a lot of issues over the years. (I'm also cynical
about trillions of bits of magnetic flux, but my experience with
that has been better).
Especially with the "cheap" stuff (or top dollar stuff that is
actually cheap stuff with a big price tag), there often seem to have
bad spots on the drives that sometimes OpenBSD doesn't handle
gracefully. Writing the entire surface of the drive seems to find
and lock out the bad spots in advance of their use for data. Ideally,
I should probably write all zeros AND all ones, but if I'm in a rush
to get something in production (or BACK into production), I just do
zeros. Writing zeros seems to help, I can't think of a case where
I can state with confidence that writing zeros and ones did something
better than just zeros.
For example, last week, I stuck a 60g USB drive on a machine, rsync'd
a bunch of data to it, and a little way in, it dropped to near zero
performance. No obvious error, but the data stopped moving and the
USB system seemed to basically stop. Could not reboot because the
OS couldn't umount the USB stick. Power cycled, dd'ed zeros over
the drive, and now I've got no issues with it.
I've been able to extend the life of flakey SSDs the same way (don't
say "write fatigue", these drives haven't had a fraction of the
writes to be worried about "write fatigue". They just weren't good
drives).
Plus...probably not a bad idea to know what data is on a USB drive
anyway.
Nick.
