[mythtv-users] Another take on large-capacity disk enclosures

[f-myth-users at media.mit.edu]

    > Date: Mon, 15 Jun 2009 15:00:36 -0700 (PDT)
    > From: "Yan Seiner" <yan at seiner.com>

    > On Mon, June 15, 2009 2:23 pm, f-myth-users at media.mit.edu wrote:

    > > The plan is to use the disks as a JBOD (and in fact most of them will
    > > be sleeping most of the time, so the two cards' performance with many
    > > disks active should be irrelevant, and means that, while a power
    > > supply that can handle the peak is advisable, it will be quite
    > > unloaded most of the time [are 80+'s still efficient with such low
    > > loads?  haven't checked recently]---I'd rather not get fancy with
    > > multiple power supplies and something to switch 'em on/off).  I'm
    > > also thinking that, unless the NAS mobo gets other duties, it might
    > > spend most of its time with its CPU throttled (and -maybe- sleeping
    > > if the stars align and that's reliable and fast to resume).

    > Practical considerations:

    > It takes a desktop-class system about 6-10 hours to check a 2 TB raid-5
    > array.  You really don't want a single large array comprising of 20 1.5 TB
    > drives; fsck would take weeks.

I probably should have mentioned (but it was already too long :) that
I wasn't planning to have a filesystem that spanned more than one disk.
(And I'm not planning on any sort of RAID at all; see comments re clone.)

I like LVM, but I don't like the idea of one spread out over so many
disks, because the failure probability (and reconstruction time) would
indeed be insane.  I'll be doing something akin to Storage Groups, e.g.,
big pile of large-ish files and I can afford a trivial level of supra-
filesystem indirection to decide which disk any given file is on.
(At worse, it's a link farm; trivial.)

    > If you pick your hardware right, you can get by much cheaper with esata +
    > port multipliers at little to no performance penalty, especially if you
    > design the wiring so that drives on the same port multiplier are in
    > different logical groups.  Port multipliers come in 1x5 versions, so 4
    > esata ports == 20 drives.  See the sata maillist for hardware that works.

That's a good plan; I hadn't checked that list's archives & will do so.
(But---assuming I don't run out and buy the TR8M on sale today---it's
not immediately clear to me where the cards/multipliers/disks are living.
And since I don't anticipate I/O to more than a couple of disks at a time,
in general...)

    > A single moderate PS will easily handle 20 drives; that's only 200W load
    > max.  Even at startup, you're only talking maybe 300W to spin up all the
    > drives.

Right.  And I'm hoping that even with 19 spun down and a motherboard
idling, the PS is still efficient enough---but that's the default case
for -most- power supplies, so I suspect I'm good... :)