UFS 'limits'?

Mark wrote:
anon wrote:
thanks, i've been perusing the opensolaris.org zfs forums
and
reading some of the Sun stuff also... i may just ask Sun,
i'd love
to see white papers that indicate they hammered a system
with a
billion files on it or something.

I saw a presentation of ZFS at the London OpenSolaris users' group a
while back, and I remember them talking about the ZFS nightly torture
test. They do something insane like create millions of files, rename
them, delete them, etc. all while modifying disk pools, creating /
breaking mirrors - all sorts of mad stuff that you'd never,ever in a
thousand years put a live production system through. And to cap it off,
they regularly simulate power faliure into the mix as well while doing
all that. From what I recall hearing, they've never lost any data or
had a corrupted filesystem in the whole time they've been running it.

Can anyone from Sun clarify or confirm that ?

-Mark

I've read the same thing. See page 32 of the LOSUG presentation:
http://www.opensolaris.org/os/commun...ff-bonwick.pdf

More good stuff at:
http://www.opensolaris.org/os/commun...presentations/

In article ,
anon wrote:
Are you really talking about 100 Million entries per
dir?

no, ~100,000,000 total files and dirs... no more than
about 45 files max within a single directory.

Then neither UFS nor ZFS have a problem.

--
(home) Jörg Schilling D-13353 Berlin
(uni)
(work) Blog: http://schily.blogspot.com/
URL: http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily

Mark wrote:
anon wrote:
thanks, i've been perusing the opensolaris.org zfs forums
and
reading some of the Sun stuff also... i may just ask Sun,
i'd love
to see white papers that indicate they hammered a system
with a
billion files on it or something.

I saw a presentation of ZFS at the London OpenSolaris users' group a
while back, and I remember them talking about the ZFS nightly torture
test. They do something insane like create millions of files, rename
them, delete them, etc. all while modifying disk pools, creating /
breaking mirrors - all sorts of mad stuff that you'd never,ever in a
thousand years put a live production system through. And to cap it off,
they regularly simulate power faliure into the mix as well while doing
all that. From what I recall hearing, they've never lost any data or
had a corrupted filesystem in the whole time they've been running it.

Can anyone from Sun clarify or confirm that ?

-Mark

I've read the same thing. See page 32 of the LOSUG presentation:
http://www.opensolaris.org/os/commun...ff-bonwick.pdf

More good stuff at:
http://www.opensolaris.org/os/commun...presentations/

Joerg Schilling wrote:
In article ,
anon wrote:
Are you really talking about 100 Million entries per
dir?

no, ~100,000,000 total files and dirs... no more than
about 45 files max within a single directory.

Then neither UFS nor ZFS have a problem.

But with UFS you maybe cannot even fsck this filesystem any more - in case
you have to (with logging very unlikely you have to fsck, but not impossible).

fsck would require several GB of memory to check this filesystem - but the
binary is still shipped as 32-bit and cannot allocate more than 3.9 GB.

With 1.5 Mio files fsck grew to ~50MB peak usage. Expect this number to
scale ~linear with the number of blocks/inodes allocated.

--
Daniel

I've read the same thing. See page 32 of the LOSUG presentation:
http://www.opensolaris.org/os/commun...ff-bonwick.pdf

I remember that now. Wow. "Probably more abuse in 20 seconds than you'd
see in a lifetime... ZFS has been subjected to over a million forced,
violent crashes without ever
losing data integrity or leaking a single block"

Just a pity I've had to deploy Solaris 10 now instead of being able to
wait for ZFS. Still, I suppose that's what Live Upgrade is for!

-Mark

(Joerg Schilling) writes:

If you go to more than 1 Million entries per dir, you need plenty of RAM
with ZFS and if you have this, UFS will not be slower.


well, depends on what you do; the degradation you are talking about is
file creation; ufs loses when you do lookups in large directories
or single file creates. If you access/create all files on one
go, then perhaps not.

Casper
--
Expressed in this posting are my opinions. They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

On Wed, 31 May 2006 20:27:10 +0000, Wes Williams wrote:

yes the world's first
128-bit filesystem

Does ZFS provide support for the "near POSIX" ACLs that Sun's UFS provides?

- Andrew

Andrew Gideon writes:

Does ZFS provide support for the "near POSIX" ACLs that Sun's UFS
provides?

ZFS uses NFSv4-style ACLs, which are a superset of POSIX:

NFSv4 introduces a powerful new ACL model. It's powerful enough
that every POSIX-draft ACL can be translated into an NFSv4
ACL. But NFSv4 ACLs can go beyond POSIX-draft semantics; thus, not
all NFSv4 ACLs can be translated into POSIX-draft ACLs.

http://blogs.sun.com/roller/page/sam...cls_everywhere

There's RFC draft that describes the mapping:

http://www.citi.umich.edu/projects/n...mapping-04.txt

--
David Magda dmagda at ee.ryerson.ca
Because the innovator has for enemies all those who have done well under
the old conditions, and lukewarm defenders in those who may do well
under the new. -- Niccolo Machiavelli, _The Prince_, Chapter VI

On 4/6/06 10:30, in article , "David Magda"
wrote:

Andrew Gideon writes:

Does ZFS provide support for the "near POSIX" ACLs that Sun's UFS
provides?

ZFS uses NFSv4-style ACLs, which are a superset of POSIX:

A presentation by Tim Graves at last year's LOSUG suggested ZFS ACLs were
"NFSv4/NT-style ACLs".

http://www.opensolaris.org/os/commun...zfs-technical-
jeff-bonwick.pdf

I don't know how different POSIX ACLs are from NT ACLs, so what impact them
being more NT-like than POSIX-like would have.

Cheers,

Chris