[NTLUG:Discuss] Software Raid/Small NAS box ideas

[Kelledin]

On Saturday 25 January 2003 07:14 pm, Fred James wrote:
> So that brings me back to the question I've always wanted to
> ask: Isn't a mirror (wherever, or however) only a possible
> protection against lost data due to a hardware failure?
> Is there any mirror that would know the difference between
> "good" and "bad" data?
> Have missed something here?

Between "good" and "bad" data?

Modern drives embed a great deal of ECC/parity information 
internally, so as long as the drive's solid-state logic is in 
order, it can reliably detect when it's having a mechanical 
failure or developing a bad sector.  The controller generally 
never sees this parity info (except, perhaps, during a low-level 
format), but it should get notified by the drive when bad 
sectors or mechanical failures occur.

This does not, however, cover a failure in the solid-state logic 
of the drive.  A drive whose solid-state logic is acting up 
could do all manner of unpredictable things, including taking an 
electrical dump on the entire HDD bus.  Fortunately, solid-state 
logic almost always outlasts mechanical parts anyways. ;)

(On a side note, some ancient hard drives didn't embed parity 
info.  To cover this, RAID2 was developed, which arranges the 
drives for Hamming ECC.  This way the controller can reliably 
detect when one drive is spitting out incorrect data, which 
drive is doing it, and what data it should be spitting out.)

On Saturday 25 January 2003 05:59 pm, Darin Smith wrote:
> Capability of supporting at least 4 drives, figuring on RAID-4 
> for redundancy and efficiency of storage volume (though not as
> performance-happy as mirroring--since it's destined to be
> networkable filesystems, I see performance as less of an issue)

First (biggest) flaw: why RAID4?  RAID5 pretty much completely 
obsoletes RAID3/4, and most any controller that supports RAID4 
also supports RAID5.  Most modern RAID controllers don't even 
bother to support RAID3/4 anymore.

Second: are you planning to do this thru software RAID?  
RAID3/4/5 takes up a lot of CPU cycles when you do it via 
software.

> Compact

Hmmm...exactly how compact?  1u compact? Pizza-box compact?  
Cigar-box compact?

Shuttle and VIA are cooperating to produce a small-form-factor PC 
that might do what you need.  Small, cool-running, 
power-efficient...I'll have to look up the exact name though.  
It runs on a VIA Eden motherboard with a fanless soldered-on VIA 
C1 CPU.

PC Power & Cooling (http://www.pcpowercooling.com/) produces 
Sleekline systems that are worth checking out.  Not sure they're 
cheap enough for you though.  People used to be able to get the 
Sleekline chassis by itself (1u pizza-box type deal), but that 
doesn't seem to be possible anymore.  I haven't done any real 
digging though...

Intel's CA810E motherboard would probably do all you need in a 
very small space.  i810 chipset (duh), socket370, onboard 
video/sound/LAN, PC100 support.  It's primarily designed for 
entry-level desktops, but it's solid enough for small servers 
too.

And if you absolutely must go even smaller, look into EBX/PC104 
single-board computers and 2.5" hard drives.  Perhaps even a 
Cappucino(sp?) bit (like our fearless leader Chris Cox has) 
would be an option, if you could go with only one drive.  That 
kind of stuff commands a high price though, and you're likely to 
have to custom-design the enclosure if you get an EBX board. ;)

> Does anyone have any opinions on AFS vs. CODA vs. NFS?

I've found that NFS is generally the best supported across all 
Unices.  Since you're going to be using Samba anyways, though, 
you might stick with just SMB--not everything works over SMB, 
but it's worth a shot...

-- 
Kelledin
"If a server crashes in a server farm and no one pings it, does 
it still cost four figures to fix?"