Re: ZFS directory breadth vs. directory depth

Tan Shao Yi <tansy@xxxxxxxxxxxxxx> wrote:

Was wondering if anyone has done an analysis of directory breadth
vs. directory depth in ZFS?

I haven't seen one, but you can make some guesses based on the
architecture.

Is there a "recommended" value for these two aspects?

I doubt recommended is the right word. Less is always faster. Like a
traditional filesystem, every directory in the path requires an
additional lookup and disk access on first use (then hopefully it's in
the directory cache and it doesn't matter as much).

Within a directory UFS and traditional filesystems require traversal of
a linear list to find an individual file. ZFS directories are ZAP
objects. At large sizes (which is the only time this would come up), a
fatzap object would be used. A fatzap object implements a 64 bit hash
of the names. So at large sizes, a linear traversal is not required and
lookup should be faster than in a UFS directory of the same size.

That said, I think locality of use comes into play here a bit. With
large memory resources, anything used recently will be in the DNLC,
reducing any issues here. If things get used once and then not again
for a long time, then the differences show up more. So that depends on
exactly how the application behaves.

We have some thoughts about using ZFS for systems such as mail spools
and squid proxy caches (the L1 and L2 values in squid); was wondering
if anyone in the group here has experience using ZFS for these
purposes?

I haven't tried it yet. I hope you'll let us know if you do some
comparisons.

You may also want to use the latest Solaris Express builds. There are
quite a few ZFS related changes between Solaris Update 2 and the current
Express version.

--
Darren Dunham ddunham@xxxxxxxx
Senior Technical Consultant TAOS http://www.taos.com/
Got some Dr Pepper? San Francisco, CA bay area
< This line left intentionally blank to confuse you. >
.