[vox-tech] RAID hardware recommendations, software support info

[Bill Broadley]

> >    - What hardware do you use? (controller, drive model/type/size)
> 
> Dell PowerEdge server (3 YO) cale with MegaRAID controller that has/had
> drivers built-into the kernel (2.4). Worked great.
> 
> I think I/We also used a qlogic RAID controller, (but I might be
> mistaken.) However, for this one, though there was a loadable module
> available from the vendor, the driver source was not built into linux.
> 
> Both were SCSI based RAID controllers, not IDE/EIDE.

Someone I know just got a large Dell raid with the PERSC raid controller,
he was VERY disappointed in the performance (under 50 MB/sec).

> >    - What software do you use to manage the RAID system?
> 
> Sofwtare? For the hardware based RAID, the controller comes with a mini
> BIOS and set of tools accessible at boot time to permit configuring
> logical volumes and RAID type/disks used for type. From here, I could set
> a hot-spare for failover support. This is where initialization and format
> for volumes was done.

I prefer not to have to reboot to play with a raid.  Ideally I never
have to physically touch it, our machine room is a loud, crowded,
uncomfortable room, I'd much rather be able to manage such things
from home, especially if it's 2am (isn't that when all server problems
happen?)

> >    - How do you know when a drive is damaged?
> 
> Hardware (only) For the ones we use(d) There is an audible beep that lets
> us know there is an issue and this is associaated with a light on the
> POwerVault storage tower (with 12-15 SCSI Drivers in hot-swappable bays)
> and the "idiot lights" shine "red" when there is a failure, steady green
> when life is good, and blink green when rebuilding. (I think this is what
> it did --- been a while since a disk failure has occured.)

Great if it's under your desk and your there, er, sub-optimial if your
not close at hand.

> As to when to swap, it is best to *not* hot-swap out a drive and then
> hot-swap in a new one when you can avoid it. It can be done, but there is
> risk. If you can afdford to schedule a down time, then that is best.

Agreed, fallover and scheduled boot is prefered.

> Then you power down the system and then pull out the bad drive, put in the
> good replacement (with the same SCSI ID as the one taken out) and on
> reboot, (most of the time) the RAID controller realizes the new drive has
> no data and rebuilds the missing data.

Ahh, this implies using non-SCA drives which I don't recommend, it's not
worth it.  SCSI-ID jumpers, seperate power cables, seperate data cables,
basically a nightmare.  I buy 5 buy enclosures, they take one power cable
(2 for fallover) and one data cable, no messing with scsi id's.

Other things to contemplate:
	If you have redunant disks do you have spare cables?
	Spare controllers?
	Spare machines?
	Spare fans?

I really like the fact that I can keep spare 39160's on other machines,
and migrate 5 disk raids between machines as necessary.  Recovering from
a dead 39160 is easy if you have another.  I've seen many sites with
fancy hardware raid controllers that are expensive enough so they don't
have spares.  If they die then your looking at waiting for fedex before
you get your disks online.

With a spare 5 disk enclosure you can migrate raidsets between machines
in a few minutes or two without tools.

-- 
Bill Broadley
Mathematics
UC Davis