Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 20 Apr 1997 12:42:20 -0700 (MST)
From:      Terry Lambert <terry@lambert.org>
To:        jbryant@tfs.net
Cc:        bakul@torrentnet.com, freebsd-hackers@FreeBSD.ORG
Subject:   Re: disklabel -- owner?
Message-ID:  <199704201942.MAA08401@phaeton.artisoft.com>
In-Reply-To: <199704201804.NAA25189@argus> from "Jim Bryant" at Apr 20, 97 01:04:41 pm

next in thread | previous in thread | raw e-mail | index | archive | help
> > - It should be able to make use of a disk database -- not disktab
> >   but something that describes the properties of various disks.
> >   Sort of like a termcap, MIB or PPD.
> 
> there is nothing broke with disktab...  what can you add that disktab
> does not already have?  heck most of the fields of disktab are not
> properly used to begin with, such important things such as latency
> usually go ignored by whoever writes disktab entries; maybe due to
> lack of info from the manufacturer...
> 
> second, never assume that i want your partition info, i may partition
> my disks differently...  such things as m-o disks to zip disks may be
> safe assumptions, but on true hard disks, i want seperate /, /tmp,
> /var, /home, /usr, and sometimes seperate /usr/local...  and the sizes
> i pick for each may indeed be different from the sizes you pick...

I partially agree.

I believe there should be no disk database whatsoever... there should
be a "rule of thumb" based on memory size to pick a default size for
swap, and another fuzzy limit for picking seperate or common root/usr,
and another for picking "additional slices".


The idea of a disktab dates back to when disk geometry tunables had
to be communicated to the kernel by the user reading a manual for the
disk, and then typing in data from that manual, and then the tunables
would actually increase performance.  Except in specific old hardware
cases, where the disk geometry can't be obtained from the disk, these
days are long gone, and best forgotton.


The sector count information should be retrieved by ioctl()'ing the
device driver to get back a count, then applying the rule set to come
up with most of what disktab/other would have had in it anyway.

Surely, there are rules of thumb for generating disktab entries, right?


In any case, the partitioning tool ought to support space allocation
based on expected needs for the selected installation type so you
don't get partially through the install, only to realize /usr is 10M
too small and / is 10M too big.  This requires resizing intended
allocations down when others go up, etc.., but it's based on sector
count more than anything, and doesn't requie moving around data on
a living FS.


> > - It should allow *moving* a partition or a slice.
> 
> good idea, why not an extend or shrink utility too?

For an existing FS?  Because you would need to re-newfs the thing,
for one, unless you rewrote JFS (I have a part of a start on that;
the PPC host environment I'm using is AIX; it's not likely to be
more than read-only and "mostly" working any time soon).  Not a
day at the beach.

Extending is both less and more problematic.  It's less because you
can pretty much safely do it, but it's more because once you do it,
you now have to unscrew cylinder group and other allocation policies
to prevent unequal fragmentation, etc..  Again, not a picnic.


> > - You may want to look at some DOS/Windows disk tools for ideas.
> 
> well...  unrm?  how about a defragger?

I think you'd probably need a "fragger" first, otherwise it will
never have anything to do and will die of boredom.  8-).


					Regards,
					Terry Lambert
					terry@lambert.org
---
Any opinions in this posting are my own and not those of my present
or previous employers.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199704201942.MAA08401>