[prev in list] [next in list] [prev in thread] [next in thread] 

List:       toasters
Subject:    RE: Vol/qtree/dir/file snapshots (was Re: DataONTAP 6.2)  ORACLE
From:       Brian Tao <taob () risc ! org>
Date:       2002-03-31 6:48:48
[Download RAW message or body]

On Thu, 28 Mar 2002, Chris Lamb wrote:
>
> Yes, it is a fundamental law of computing that files and databases
> will always expand to fill the available disk space, but why is it
> that you can't even buy a 9GB drive anymore?  A 20-30GB database,
> even if you double or triple it in size, is still going to fit on
> *one* shelf with 9- or 18GB drives.

    Going *way* out into fantasyland here: what we need are
PCMCIA-sized 1-inch drives.  Not like the kind you see today... these
ones would have a single, tiny platter spinning at 30000 rpm with an
FC-AL interface.  A 2U-tall enclosure could fit 72 of those across a
19" face with some room to spare.  Break that up into 5 14-disk RAID
groups and 2 hot spares.  If each disk is 4GB, you end up with about
230GB of useable space.  Not very dense, but super-high spindle speeds
and lots of them.  Lots of blinkenlights too.  ;-)

> A long time ago I was going to recommend that filers come with a
> pair of internal boot drives - like a mirrored pair of 4GB or 9GB
> drives _strictly_ for use as the boot volume, with some space for
> logs, etc.

    You'd lose the ability to cluster then, since the partner will not
have access to those drives.  I suppose you could have internal
mirrored drives (plus a spare), and split them like the NVRAM.
Writes would then be mirrored to both local and partner drives.
However, that's a backwards step from eventualy filer/storage
virtualization.  It doesn't make much sense to have significant
storage physically tied to a filer head when you move to N+1
clustering, a SAN fabric between filer heads and the drive pool, etc.
Being able to boot off of a flash image (like on the F880 and newer
models) seems like the right way to go though.
-- 
Brian Tao (BT300, taob@risc.org)
"Though this be madness, yet there is method in't"


[prev in list] [next in list] [prev in thread] [next in thread]