ZFS, OpenSolaris, Nexenta -- eating my words

Michael Halcrow mike at halcrow.us
Sat Nov 19 11:39:13 MST 2005


On Thu, Nov 17, 2005 at 11:05:17PM -0700, Michael Torrie wrote:
> Well, that has suddenly changed with the announcement of Sun's
> uber-fs, ZFS [1].  I only have just a small inkling of what ZFS is
> all about, and I get the feeling this is really big.  Actually huge.
> This is a major advancement in free unix distros.  In my mind this
> is as huge as xen is going to be.

That's some pretty good Kool-Aid, ain't it, Michael? :-)

http://marc.theaimsgroup.com/?l=linux-kernel&m=113242108929933&w=2

And the PDF presentation on the page you linked to reeks of marketing
hype:

http://www.opensolaris.org/os/community/zfs/docs/zfs_last.pdf

Slide 19 is just ridiculous. They totally misapplied Moore's Law to
justify their 128-bit address space design decision; those who
actually read Moore's paper would know that Moore's Law actually
refers to the density of functional components on silicon devices, not
the density of magnetic domains on hard disk platters. They are
entirely different technologies with entirely different trends.

Incurring the overhead of a 128-bit address space and justifying it by
saying that it will be needed a couple of decades from now is the same
class of fallacy as incurring the cost of a 512-bit symmetric key
while making wild conjectures about computational power a couple of
decades from now. That ``exceeds quantum limit of earth-based
storage'' claim is pure marketing babblespeak that has no bearing on
real-life problems in the marketplace, like those caused by eating up
a whole 128 bits for every address you need to pass around in ZFS.

It looks like ZFS took a page out of a book on good database design
and processor architecture design, and then they mushed them together
and called it a filesystem. How much more useful is ZFS over our
current technologies, that we should throw it all away? As long as we
have to throw away our traditional notions of block devices, inodes,
dentries, and so on (which, by the way, is not a unique concept by any
means -- see Object Stores by IBM Haifa labs
<http://www.haifa.il.ibm.com/projects/storage/objectstore/>), why not
just go all the way and extend Reiser4 with the volume-less
capability?

I agree that we could do better, and ZFS looks like a step in the
right direction, but I would be skeptical of any suggestion that it is
the best alternative we have at the moment.

Mike
(Disclaimer: I speak for myself and not necessarily my employer.)
.___________________________________________________________________.
                         Michael A. Halcrow                          
       Security Software Engineer, IBM Linux Technology Center       
GnuPG Fingerprint: 419C 5B1E 948A FA73 A54C  20F5 DB40 8531 6DCA 8769

A hacker does for love what others would not do for money. 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 481 bytes
Desc: Digital signature
Url : http://plug.org/pipermail/plug/attachments/20051119/976835d6/attachment.bin 


More information about the PLUG mailing list