[Novalug] best hard disk setup for home file server?

[Bryan J. Smith]

First off, the newer MD meta removed some long-standing issues
with some programs believing it was just an Ext2/3 slice (partition).
Using the AutoRAID partition id also helps as well.  That was a long
standing issue that has been better mitigated.

Secondly, the multiple layers in the kernel with MD + LVM + DM
features can result in a race condition.  It's been mitigated more and
more, but I know I'm not alone in avoiding putting LVM atop of MD.

I almost always use hardware RAID (and that's _real_ hardware RAID
with an ASIC/microcontroller, like AMCC/3Ware PPC400 or Intel/Marvell
XScale, etc...), and a trusted hardware RAID solution with long-standing,
forward volume support (e.g., 3Ware's dozen years of volumes being
forward compatible with their newer cards).  Better yet is to use a RAID
vendor that has an open meta, so DeviceMapper (DM) can read it without
the hardware RAID card (again, a 3Ware bias is showing, although
anything prior to 9550SX is a slouch for RAID-5/6, and only good for
RAID-1 as they are "storage switch" ASICs).

Alternatively I used to put MD atop of LVM.  This "protects" the MD
meta from being modified because nothing touches LVM as it looks
nothing like Ext2/3.  It also "localizes" MD corruptions to only the
filesystem being used instead of all filesystems (LVM atop of MD).

LVM also adds many more mapping options thanx to DeviceMapper
(DM) support.  And DM can use FRAID (fake RAID) volumes which
gives you seemless boot support, while still being readable.  More
recently LVM has added DM with mirroring (RAID-1) in many
implementations, removing the need for MD.  Being LVM+DM with
mirroring built-in, it's one volume set for both capabilities.  It has a
number of sanity checks.

The sooner we move to a single volume set manager, the better.  The
layered approach can result in race conditions, seen it too many times.

As far as 2.5", cooling and power are major benefits, along with portability.
A lot of people try to say that 2.5" is "no better" than 3.5" when it comes
to MTBF/AFR, but looking at the "operational parameters," 2.5" can take
a crapload more shock and other tolerances.  ;)  A lot of people go ARM
or Atom (although Intel is often "recycling" 90-120nm fabbed 965+ICH7
chipsets that suck 22+3W) for power.  Beyond that, even 10-15Krpm
drives are all 2.5" and have been for years (the only exception I know of
is Seagate Cheetah, but if you open it up it will be 2.5-2.6" platters and
the 3.5" is largely for the 25mm height and heat sink).

More personally, I always buy the latest 2.5" drives for my portables, and
recycle my old disks to my servers.  E.g., I just upgraded to 2x640GB in
my Gateway P-7811FX and my wife's notebook, so 4x500GB are being
recycled into my server, and 8x250GB are being recycled into secondary
servers (my Florida home, my parents small business and home server).

I can say without hesitation that 10 out of 10 times when servers are
randomly crashing, removing the external USB/FireWire storage solves
the problem.  And those weren't even volumes in most cases, but just
the "backup" solution.  I only use hot-swap 2.5" drives and have had
that attitude since 2006.  Correspondingly, I haven't installed a server
with anything but 2.5" drives since 2006 as well.

About the only time I've used 3.5" drives in the last 3 years is for near-line
MAID, and they are commodity and not considered production, only for
near-line, temporary storage.


----- Original Message ----

From: Richard Ertel <richard.ertel at gmail.com>
@Bryan:
RAID is being done via MD, yes. And LVM is on top of MD in this case.
Have a short reason why that's not recommended? I used LVM in the past
w/o RAID, but decided that I wanted some amount of fault tolerance.

What is your definition of system disks? You mean / and /home and
such? I should have clarified that the RAID + LVM stuff is just file
storage only (/storage to be exact). That being said, I do have plans
to switch / and /home over to RAID-1 on their own internal drives.

Is there any advantage to 2.5" drives over 3.5" drives besides size?
size isn't an issue for me, as i have these 4 drives in a big server
case in my unfinished storage space. i have 2 drives in a 3x5.25" fan
bay that can hold 4 drives, i'm just giving them a bit more room. so
my 4 drives are being cooled by two 120mm fans.

my initial curiosity in external usb drives is simply because i favor
modular systems in general, to an extent. i predict troubleshooting
and repairing would be easier with external usb drives, but not worth
it if the reliability isn't there.