[KLUG Members] Re: RAID

[Bryan J. Smith]

--=-wFRhpwHDYp848uBHaO64
Content-Type: text/plain
Content-Transfer-Encoding: quoted-printable

On Sun, 2002-12-15 at 00:27, Adam Williams wrote:
> If your going to use hardware RAID use a card ***NOT***
> integrated on the mobo, and have another one around.

I think what Adam meant by "use a card" he meant "use a _real_ hardware
card" and not just a card with an ATA controller, a BIOS and nothing
else.

Otherwise, in case you didn't know Adam, mainboard ATA RAID is just an
extra ATA controller and BIOS, glued via PCI logic, just like any PCI
card with an ATA controller and BIOS.

So if you get a mainboard with two extra ATA channels thanx to a Promise
controller and their proprietary RAID BIOS, it's the same thing as a
FastTrak card.  Same thing with HPT, etc...

> Thats another reason why I like the IBM ServeRAID SCSI RAID cards,
> they are plentiful.

IBM, DPT (now owned by Adaptec), Adaptec and most other SCSI RAID
controllers (as well as the Promise "SuperTrak" and Adaptec 2400A ATA
RAID controllers, which are based off of traditional SCSI RAID design)
use an Intel i960 microcontroller.  Top-end Mylex (owned by IBM), at
least one Adaptec model, and a few others use StrongARM/XScale
microcontrollers.  The microcontroller has its own memory (EDO or
SDRAM), embedded OS, drives the ATA controllers and handles all
transfers between the rest of the system, over the PCI bus, and all
on-card components.

3Ware uses a slightly modified approach.  They use an Application
Specific Integrated Circuit that arbitrates transfers between the PCI
bus (64-bit in the 7000+ series), the ATA channels and 0 latency Static
RAM (SRAM).  This is not unlike a CPU cache controller would be
designed, or an advanced network switch fabric.  The reason is that
3Ware realized the advantage of ATA is its non-blocking I/O design.  ATA
is "dumb" block I/O, that relies on quick, direct transfers, whereas
SCSI is intelligent, command bus that is far better for lots of command
queuing.

Using microcontrollers and latency-ridden DRAM technologies introduce
cycles of I/O delays that kill ATA's advantages.  In the case of RAID-5,
you're already introducing latecy with XOR operations, but in RAID-0, 1
or 0+1 (aka 10), non-blocking I/O far better.  The disadvantage of the
ASIC+SRAM approach is that SRAM logic is massive and expensive compared
to DRAM (Tech: SRAM logic is traditional boolean design, whereas DRAM
logic is simple-cell which has "leakage").  So 3Ware controllers only
feature 1-4MB of SRAM, compared to 16-128MB EDO/SDRAM found on
microcontroller+DRAM designs.  So RAID-5 write performance, especially
lots of random writes, is less than optimal.

> Of course, you could always keep a spare mobo around.... but
> I can always find other uses for my home system dollars.

Well, if you're already spending a lot on the disks, especially for
SCSI, then the controller cost is minimal.  But you should _always_ make
sure you can read your volumes somewhere in the case of controller
failure.

In the case of 3Ware, I've been using their products since their initial
5000 series.  I have moved drives between 5000, 6000 and 7000 series
without loss (assuming the firmware is the same/equivalent or newer).

Which is my last point.  It doesn't matter what hardware RAID solution
you use (I've used 3Ware, Adaptec, DPT, ICP-Vortex, Mylex and others),
you _must_ make sure your firmware and driver combination are
_compatible_ with each other.  That means if the driver is in your
kernel, you need to make sure you don't have an older firmware that
isn't compatible with the newer driver.

> I've found software RAID in Linux to perform very well,   recovery is
> easy; assuming you have some idea what was in /etc/raidtab (nothing like
> a hard copy :).  It even supports hot spares.  It is certainly
> sufficient for the home/small-office user.

Software RAID of the OS is very portable between controllers, and you
don't have to worry about firmware-driver compatibility issues.  Of
course, you must make sure you don't move the disks to a system with a
kernel with an older MD/LVM driver either.

--=20
Bryan J. Smith, E.I. (BSECE)       Contact Info:  http://thebs.org
[ http://thebs.org/files/resume/BryanJonSmith_certifications.pdf ]
------------------------------------------------------------------
  The more government chooses for you, the less freedom you have.


--=-wFRhpwHDYp848uBHaO64
Content-Type: application/pgp-signature; name=signature.asc
Content-Description: This is a digitally signed message part

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.7 (GNU/Linux)

iD8DBQA9/CbuDjEszaVrzmQRAuB7AJ9BZxAcdaDZhnxlxxyAS6eG932ZfwCfZs3c
M4BEe4xSQSLiS6ClPnEFWE0=
=ThyG
-----END PGP SIGNATURE-----

--=-wFRhpwHDYp848uBHaO64--