Re: Logical volume management

In message <20051118183115.V52197_at_fledge.watson.org>, Robert Watson writes:

>However, as recent discussion of ZFS illustrates, you can get some really 
>nice benefits from tight integration -- logical file system volumes that 
>offer layout-based quotas, reliability or performance properties based on 
>location in the name space, etc.

I can't help wonder if ZFS is going trigger part of the UNIX crowd
into renouncing the sacred teachings of Software Tools as "without
contemporary relevance" and move to have the edited out of the
official records.

It is of course an extraordinary claim that a concept invented over
35 years ago should still be verbatim respected today as the one given
truth, and clearly other metaphors for computing, the desktop most
notably, have made a compelling case for their acceptance.

But at least I still belive firmly in the software tools paradigm
and at the same time I will readily admit that we in the UNIX crowd
have misappropriated our heritage somewhat (sockets not named in
filesystem, SVID IPC, etc).

But I find that ZFS grinds my sensibilities, it sounds to me like
it is trying to solve too many problems by lumping them together,
leaving us with a non-intuitive conceptual model which will, take
its toll in mistakes and oversights and in due time, have to be
reverted.

UNIX has never really managed to make mounts a dynamic thing, they
happen at boot and are nothing but trouble after that.  It is
therefore understandable that one would be tempted to push a load
of desirable functionality into one filesystem and in that act paste
over the fact that the harder problem of truly making filesystems
a comodity concept in UNIX is still unsolved.

My experience so far has been that such circumlocation of features
is nothing but trouble later on, when outside pressure forces things
to be done right before new desirable functionality can be enabled.

Failing to address the problem of proper level of abstraction and
proper boundaries for feature addition, will eventually make any
operating system a mess of separate and uncooperative features of
little actual worth to the community.

So while I think porting ZFS to FreeBSD is a worthwile exercise, I
would urge anybody really interested in pushing us forward to
critically consider mountpoints, and see if the conceptually simple
model of storage management they offer cannot be retained or even
improved, while gaining some of the benefits which ZFS allegedly
brings.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk_at_FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.