On 05.06.2014 11:12, Michael Tokarev wrote:
04.06.2014 18:00, ronnie sahlberg wrote:
That would mean you get to use the 10 version of the cdb even for very
large devices (as long as the IO is for blocks at the beginning of the
device) and thus provide partial avoidance of this issue for those
large devices.
That may make some bugs "ghosty", so to say. Ie, if there's a bug in/with
16 version of a command, you'll hit it only when you actually try to access
a "far" area of a drive. Which means you're unlikely to hit it while trying
to reproduce in a clean environment, even after using a large device. Or,
the bug will be triggered at random, since data placement on the filesystem
is effectively (from user PoV) random.
To my taste it is better to make it a bit more deterministic.
BTW, while debugging a case with a bigger storage supplier I found
that open-iscsi seems to do exactly this undeterministic behaviour.
I have a 3TB LUN. If I access < 2TB sectors it uses READ10/WRITE10 and
if I go beyond 2TB it changes to READ16/WRITE16.
Peter
