[svlug] New server plans moving forward

Rick Moen rick at linuxmafia.com
Sun Jan 21 17:02:46 PST 2007

Quoting Daniel Gimpelevich (daniel at gimpelevich.san-francisco.ca.us):

> > SCSI Channel A is LVD (low-voltage-differential?) and is the external JBOD
> I believe that it would probably technically be termed an external RAID
> device, regardless of any RAID capability it may or may not have
> internally.

The VA Linux Systems model 9008 external drive enclosure is termed a
"JBOD" unit because, in itself, it lacks any sort of RAID ciruitry,
having only an SCA backplane and dual PSUs to support any drives you
slide into the eight front-loading bays.  Thus, per se, it's "just a
bunch of disks" (JBOD) to the hosting system.

The fact that we _happen_ to have implemented some mirroring and
striping using the Linux md driver is technically irrelevant to the
question of what you call the external box.

As is typical for Linux servers, our Internet presence has so little
need for CPU power that the RAID overhead really won't matter at all, by
the way -- and wouldn't using RAID5, either (which was our original
intent for the JBOD, to maximise usable space).

> > - they're a RAID1+0 setup - a raid1 stripe joining 4 raid0 mirrored
> > pairs into one big huge ~144Gb virtual drive.
> It's a raid0 stripe joining 4 raid1 mirrored pairs.

Not a complaint, but just so people know:  A RAID0 striped pair has the
property of negative redundancy.  That is, while in contrast a mirrored
pair can survive the demise of either half, a striped pair is at the
mercy of the health of _both_ halves.

The usual reason one would want to stripe, regardless of that redundancy
drawback, is the advantage of being about to form a single huge volume
from both drives, plus lower effective seek times on that volume.

You should accordingly never entrust to striped storage any files you
can't afford to lose / revert to backup -- or at least be aware that
they're at considerable risk. be aware that they're at considerable
risk. be aware that they're at considerable risk.

> > AFAIK 36GB SCA SCSI drives are pretty abundant.  I'm sure hoping that's
> > true, since we now need at least one, to replace the one that failed. 
> > If we can get a few more extras as spares, that'd be great. So, if you
> > have any old 36GB drives laying around, PLEASE let me know, so I can
> > twist your arm till you donate them. :-)))
> Don't forget to mention that they must be SCA.

If I might make a suggestion:  Little SCA-to-68pin converter dongles _are_
cheap and easily found.  Ergo, it's pretty easy and routine to use them
to turn a regular 68pin non-SCA to an SCA one, that way.  So, please
make sure people understand that _either_ type of 36GB SCSI drive is
welcome.  High-voltage differential SCSI (thankfully exotic) is the 
only type we can't use.

By the way, Paul, if you haven't yet thrown away those "failed" drives, 
please hand them to me for re-checking before discarding (especially if
any of them is the 18GB one I donated to the effort  ;->  ).

> > We need to investigate mdadm's monitoring capability, and add something
> > to rc2.d to autorun monitoring, so we can get emails if/when something
> > significant like a drive failure happens to the various RAID instances. 
> > Maybe setup a mailinglist for "sysadmins" that we can have software send
> > emails to?
> That may already be in Debian. It's easy to check.

It is.

More information about the svlug mailing list