>>>>> "b" == Blake <blake.ir...@gmail.com> writes:
b> ZFS can preserve data all day long, but that doesn't help much
b> if the controller misbehaves
the most common kind of controller, or rather driver, misbehavior is
to time out commands to failed drives while ZFS waits inappropriately
long, or to freeze access to drives other than the drive that failed,
or to fail to support hot-swap properly. These problems affect
availability (no interruptions in data access), not reliability (once
put there, stuff stays there, but maybe only readable after manual
intervention). If the box is doing a ~throwaway job like tape
emulation for backup, good reliability with crap availability may be
more acceptable than the same situation on a production box.
b> it's happened to me with whitebox hardware).
and also with X4500, taking 1+ years to not completely fix, and
closed-source driver so you cannot try to fix it yourself. though
there seem to be fewer complaints with the new LSI-based controllers.
Paying a premium for good integration makes more sense if the
integration was actually good in the past. I'd also pay a premium for
something like VA Linux which bundled hardware with stable revisions
of good-quality open-source drivers, but I guess that didn't work out
well for VA.
It makes some sense but less sense to pay a premium for a situation in
which the integration could theoretically be good even though it
hasn't been in the past, because there may be FUD problems that were
solved (both presently and in the past) without your hearing about
them so you are actually getting something for the premium you've
paid, but $vendor cannot tell you exactly what you are getting because
knowing what problems were fixed would make it easier for you to fix
them yourself without paying. That's a pretty damn cynical outlook on
the situation, but it seems to be a realistic/common one.
That said why fuss about drive cost? If the drive-to-chassis lock-in
is effective, then just accept the tying as done and compare cost/TB
with Xtore or Dell FC or whatever. The real interop/tying thing I see
needing untangling in ZFS is for FC and iSCSI to work well---if this
were done then ZFS could be combined with a variety of competing
physical storage, and it's reasonable to expect it be well-integrated
with non-Sun FC/iSCSI---I think the claim ``well there is <FUD> so you
have to buy Sun FC targets'' will not fly so well in that case because
you are already paying an integration premium to both vendors and
might reasonably expect some combinations of them to claim and deliver
good interoperability.
pgp2XIRwL36hQ.pgp
Description: PGP signature
_______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
