Whoa, we suddenly banged the horns... I don't want this to be a
conflict, but rather a respectable constructive discussion :)
Post by Darren ReedYes. There are commercial grade file servers and file systems (NetApp &
WAFL) that come to mind. EMC, IBM and others also probably have
something to offer in that space.
Well, ZFS marketing (including from Sun and Oracle) sets presence of
checksums as an important differentiator between ZFS and other storage
solutions. And as that FUD-talk said, the "simple washing powder" (I
mean classic RAID systems) is only good if the whole disk kicks the
bucket, or at least a sector read does return a HW error. Otherwise
this turns into a guess-game of which copy (or RAID reconstruction)
of the data to trust in case of discrepancies that are not accompanied
by HW errors.
I do no know how other vendors solve this problem, and if they do.
I did have simple RAIDs go astray (Linux md and mirroring on the
motherboard, maybe Solaris SVM - though not so sure) when both halves
of the mirror were readable but contained different data. This led to
the reads either returning invalid (randomly chosen) data half of the
time, or (during resync) one disk arbitrarily being chosen as the
master copy to overwrite the other disk. So, at least for some cases,
the ZFS marketing's description of "classic RAID" was correct - it
does have the problem with hidden data rot without diagnosed errors.
DO, to your knowledge, any other solutions provide protection against
such class of situations (they do happen, verified)? Or do they just
rely on expensive hardware that is less likely to break or lie - with
no means to verify that it did not?
Post by Darren Reed"... backed by *their* paying customers -"
Yes, you are right, I missed it... but...
Whose customer one is today - this is not necessarily permanent.
Some want to hop onboard, others want to jump ship - and OpenZFS
might be the boat they'd sail next. So in a year or two they might
be (free or paying) customers of one of the illumos/OpenZFS vendors.
If they bring along money, or code, or word-spreading efforts - so
much the better.
Post by Darren ReedSo let me get this straight. Because the new owner of Solaris makes a
good profit, whatever antisocial behaviour they exhibit towards open
source communities is justified and should be used as a role model? Are
you serious?
No. Role-modeling - hell no, and I am sorry if anything I said could
be interpreted that I did imply that :)
Tracking what they do, just like being aware of any other interesting
developments in the storage industry or things related to ZFS, may lead
to implementing something that is desired by our potential customers.
Say, some of this list member's track what new optimizations come up
with cryptography and particular algorithms, and some of these make way
into OpenZFS for checksums, to make it faster.
Taking the offer to ignore Oracle to the extreme: would you propose,
alternately, that we hide in a shell and only implement things that
we thought up and believe are cool, regardless of whether anyone
else wants them, or that our users do want something else too?
There are likely things in Solaris ZFS that are irrelevant to the
OpenZFS community in general (like, say, integration with OracleDB),
and there are things which (or similar to which) we might welcome
in the open.
Post by Darren ReedWouldn't it be better to behave as they do and base the features
supported on the number of requests for them rather than simply copying
what someone else does in the hope that you can ride the other's coat tails?
There's time for both strategies. I don't advocate for only copying,
nor for only making stuff based on RFE votes (say, Saso makes a lot
of great stuff without an immediate commercial gain, AFAIK, just as
an enthusiast; while up-streamed bugfixes from companies originate
mostly as fixes to things that plagued their paying customers).
Post by Darren ReedSurely it makes sense to tend to your own flock first and make sure that
it is happy, yes?
And if your flock wants what those guys behind the fence have got?..
Post by Darren ReedOn the other hand, if the end game for companies such as Delphix & other
illumos/opensolaris spinoffs is to be bought back and folded in then
sure, copying those other features is important for the spinoffs.
I hope this is not The Plan :)
//Jim