On Wed, 2006-03-01 at 23:34, Dan Price wrote:
> Peter's mail about fsstat kstats inspired me to write up some thoughts
> I have about fsstat.  I've been playing with it over the weekend and
> some more today.
> 
> 1. Thank God (or thank Rich) that we finally have something in this
>    space.  It's a vast vast improvement.

Absolutely...

> 2. I'd like to see many more ways to slice the data:
> 
>         fsstat -A       Show all mount points for all FS's
>         fsstat -z       Skip zero'd entries (like iostat -xz)

Definitely like the -z option. It's got to the point where
I only run iostat with -z.

>         fsstat -a -F <fs>  Show all mount points for this FS type
> 
>         fsstat -a -l    Show only local filesystems (no NFS, etc).
>         fsstat -a -s    Show only non-synthetic filesystems
>                         (no lofs, mntfs, autofs)

Hm. There's this discussion over on zfs-discuss (the df output
with lots of zfs filesystems thread) which has some relevance,
in the sense that I think that df and fsstat shouldn't come
up with different ways to address the same problem.

In my case, what I'm really after (in both cases) is to say
"watch ufs and tmpfs" or "watch everything except zfs". So a
possible -X to exclude a filesystem type, and allow -F and -X
to be specified multiple times.

>         fsstat -Z       Aggregate each zone's FS activity
>         fsstat -z       Watch a particular zone
>                         (of course I'm proposing two -z's... we'll
>                          need to work that out :p)

How would this work in practice (without zfs)? I presume
you would go through the list of mounted filesystems
and pick up the zone assignment from the options? The problem
being all these lofs mounts. For example, on my machine the
root filesystem of the zones is the same device as the root
filesystem of the global zone, and the same for /usr etc - how
is the FS activity to be separated? (With zfs it would be easier
as you're more likely to create a new FS and delegate it to a zone.)

> 3. I find the output on the challeging side to visually grok:
...
>    Where does one second end and the next begin?  Some spacing or repeating
>    the titles or something would help a lot.  mpstat and friends all do that;
>    this code seems to do it on every third iteration, which I think isn't
>    that useful.

I agree with the conclusion elsewhere - with multiple lines of
output, print a header every time. Else every 20 or so (vmstat
and iostat seem to be every 18 or 19).

> 4. I'm not sure that the default output is the one which most admins
>    would want at a glance.  Just my $0.0002.

I would agree that the default output isn't obviously the one I
want. But I'm not sure that I can come up with anything better.

There's so much information here that it may be necessary to follow
the example of ps and have a '-o format' option.

> Thanks again for a very useful tool.

+1 (at least)

-- 
-Peter Tribble
L.I.S., University of Hertfordshire - http://www.herts.ac.uk/
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/



Reply via email to