[gentoo-amd64] Re: Re: Re: Re: Giving up 64 platform

[Duncan]

Boyd Stephen Smith Jr. posted <200604252300.53210.bss03@××××××××××.net>,
excerpted below,  on Tue, 25 Apr 2006 23:00:47 -0500:

> On Tuesday 25 April 2006 20:53, Duncan <1i5t5.duncan@×××.net> wrote about
> '[gentoo-amd64]  Re: Re: Re: Giving up 64 platform':
>> Of course, folks like me would then want to rearrange the partitions a
>> bit, but given today's 200 gig plus hard drives, copying a few
>> partitions around is easier than setting up a new installation from
>> stage-X.
>
> Don't use partitions! Use LVM!
>
> Okay, sure, you still need a partition for /boot (and then a partition for
> the rest of the drive, which is controlled by LVM), but that's only on one
> drive.

Well, I am using LVM, but not for / either, because that seriously
complicates the boot process.  I have / (and my rootbak partition as well)
on kernel md RAID (partitioned RAID, actually), which is acceptable, as
the kernel itself (with a few added command line parameters, if necessary)
handles md/RAID.  However, the kernel can't yet mount LVMs directly,
without help from userspace, which makes things MUCH more complicated as
that requires an initramfs/initrd, if / is LVM managed.

Layout here:

4-SATA-hard-drive RAID, with each physical drive identically partitioned.

The first partition on each drive is an unpartitioned  RAID-1 for /boot,
which because RAID-1 is mirroring, means /boot is mirrored across all four
drives and after installing grub to the boot sector on each one, means I
can boot grub from any of them, independent of the others.

Partition two on each drive is part of a partitioned RAID-6, so any two
of the four drives can drop out without disruption.  The first two
partitions on the RAID-6 are / and rootbak, identically sized and
containing /usr and /var (minus selected subdirs of each), such that the
OS image on each is self-consistent with the portage database in
/var/db/portage on each. (I learned that lesson the hard way, when I had
/usr on a separate partition and lost it, and the backup /usr was old, so
the portage database was tracking the wrong versions of everything
installed to /usr, saying they were all upto date when they weren't.)  The
third partition on the RAID-6 is LVM, which contains logical volumes for
the packages tree (by default /usr/portage/packages, IIRC, with
FEATURES=buildpkg) and it's backup image, /home and it's backup, a mail
volume and backup, a news volume (no backup), a media volume and backup, a
log  volume (/var/log, no backup), etc.  The arrangement is such that LVM
need not be started until after / (or the backup root, if I have problems)
is mounted, so no initramfs needed.  Further, should there ever be a
problem with LVM, I can at least boot to a functional root filesystem for
maintenance -- I appreciated the wisdom of this recently, as there were
some problems with ~arch LVM updates, causing folks with root on LVM to
have problems that either woudnn't touch me or would be far less severe. 
(Again, note that RAID is kernel dependant only, so as long as I keep a
known working RAID kernel around, no problems getting /root mounted.  / on
LVM is an entirely different story!)

The third partition on each physical drive is my swap partition, 4 gig
each drive, 16 gig total.  With swap priority specifically set equal for
all four, the kernel manages them as striped swap, and yes, that /does/
make swap /much/ faster!  (Not that I use it since I upgraded to 8 gig of
RAM anyway, but before that, when I had only a gig of real RAM, I
definitely appreciated the performance benefits of striped swap.)

Partition 4 on each drive is reserved as the extended partition, in case I
have need to add more than the one additional partition I actually use. 
That makes the additional partition #5, partitioned RAID-0, striped across
the four drives for speed at the cost of lack of redundancy.  Until I
upgraded to 8 gig RAM, my /tmp was the first RAID-0 partition.  It now
remains unmounted by default, as my /tmp is now tmpfs based (limited to 6
gig max, but I don't believe I've ever reached that, or caused swapping of
any kind, for that matter, I've had 6 or 7 gig of the 8 gig in use at
times, however, mostly as cache, which tmpfs registers as until it starts
swapping, as well).  The second partition on the RAID-0 contains three
subdirs, each representing some place in the system containing data that's
either easily recovered or is easily rebuilt.  p contains the portage tree
(default /usr/portage).  Of course, that's easily redownloadable, in the
event of a disk malfunction that would take out the RAID-0.  Likewise,
usrsrc is symlinked from /usr/src, and contains the various kernel trees
I'm currently using and since my last cleanup, plus a tarballs dir
containing the kernel tarballs.  Again, this data is easily
redownloadable, should the RAID-0 fail, so lack of redundancy isn't an
issue.  The third subdir on RAID-0's partition two is ccache, nice, but
rebuildable from scratch if necessary.  It's a very nice coincidence that
all three of these happen to be performance critical as well, particularly
the portage tree and ccache, and striped RAID-0 optimizes performance, at
the cost of data redundancy safety, which isn't needed for these anyway.

Even allowing for plenty of spare room on all the RAIDs and each RAID
partition, with the disks being 300 gig each, I've a bit over 100 gig left
entirely unpartitioned on each one.  That gives me plenty of room to move
partitions around and/or expand them, if necessary.  Naturally, if I need
to expand the LVM beyond its current bounds, that will be easiest.  Just
add a new set of RAID-6 partitions and add it to the LVM.  However, with
the area unpartitioned, it remains flexible enough to allow moving the
other partitions around if necessary, or to add more.  Further, while I
have well over 16 data volumes (partitions or logical volumes), nothing is
getting close to the limit of 16 partitions per SCSI host imposed by the 
kernel device numbering, or the similar limit on RAID (I don't know if 
it's 16 or higher, but it's not lower).  Back on IDE/PATA, the limit is 64
partitions.  I think I had 20-some at one point, before I switched to
RAID-0/1/6+LVM2-on-4-300G-SATA-disk.  I'm glad I made the switch.  It has
been well worth it!

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman in
http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html


-- 
gentoo-amd64@g.o mailing list