>Date: Mon, 30 Sep 1996 18:46:08 -0400
>To: James Graham <greywolf@siva.captech.com>
>From: "Kevin P. Neal" <kpneal@pobox.com>
>Subject: Re: VPS mailing list, BSD interest?
>Cc: hackers@freebsd.org, tech@openbsd.org, current-users@netbsd.org

Since this message is only marginally useful to the lists its on,
and really useful to vps-devel, I'm setting the Reply-To to vps-devel.

>At 12:24 PM 9/30/96 -0700, James Graham wrote:
....

>Note that JFS figures into this somehow, and I'm not very clear on this
>(Terry?).

I believe that jfs happens to be AIX's fs of choice, so it's not
inherantly necessary to have it too. That said, I would personally
prefer to have a journaling file system, one based on a slightly better
performer than generic ffs. (I was actually looking at getting the
Viva File system going on the Lite-based systems, and then adding
journaling and live expansion to it (It's number three on my to-do
list). Check out ftp://al.cs.engr.uky.edu/cs/techreports/225-93.ps.Z
for info on VIFS.)

>I don't know how FFS or ext2fs will fit into it, or if they will.

I seem to remember that "der Mouse" had a program to extend ffs level
2 filesystems, although it met with some serious reservations. Perhaps
with more work it could be used on non-live filesystems. I don't know
how easy it'd be to resize ext2fs.

>LFS? (Terry?). Is Margo Seltzer around? (Would she be able to contribute
>any ideas?) Her web pages looked cool (I love web pages with white papers
>online).

If LFS ever gets fixed, this would be a good candidate for extendability,
since all LFS is in the end is one large open space. Of course, the devil
is in the details. :)

>If anybody thinks this is a good idea, but doesn't have time, at least let
>me know that somebody else thinks this is neat stuff.

I for one am very glad to see this, because it is number two on my
to-do list. (Number one is to get enough disk space to do the rest
of the list. ;) As soon as my finances allow a NetBSD source tree,
I will be happy to throw what personal time I can in on this.

Having seen both AIX's and HP-UX's lvm implementations, there's a few
implementation ideas I'd like to share:

1> Extendability should be transparent.
AIX does this one right IMHO, allowing for live expansion of file
systems, HP-UX (at least in 10.01) made you unmount the filesystem.
If someone can engineer one of the file systems to live through
being shrunk, either on-the-fly or while inactive, even better
(but notice I'm not volunteering :).

2> Movable extents.
HP-UX LVM (at least) allows one to move all of the extents (*) on
one or more physical partitions (no matter how fragmented) onto
different partitions while live (leaving the new arrangement
physically contiguous). Since this could be real-time work, it'd
make moving data between disks even more convenient.

3> Sparse list implementations
One flaw I've seen in HP-UX LVM is that the logical extent to physical
extent mapping appears to be in array form, which is IMHO too wasteful
of kernal memory. Instead, I'd like to see most of the lists in the
kernal be sparse.

4> Some RAID functionality
I'm very skittish about RAID 5 in software (it's slow enough in
hardware), but allowing striping across physical media and/or
mirroring across physical media in software is IMHO a Good Idea when
used wisely. Certainly a 2 controller machine with disks on each
can manage some nice speed increases when the partition is spread
across the controlers.

David

--
David W. Rankin, Jr.  External Email: david_rankin@vnet.ibm.com
   Internal Email: drankin@davidr.lexington.ibm.com or lexgate(drankin)
      Phone: (IBM Internal) 8-545-1457   (External) 1-606-243-1457
         Alpha Pager: 1-606-259-6309