XFX RAID-XL (was "RE: [Pc_Support] XFX DoA Issues")

[Bryan J. Smith]

On Sun, 2005-07-31 at 17:26 -0500, Bryan J. Smith wrote:
> Haven't had time to mess with it since February.  Supposedly the kernel
> supports it better now.

Just checked, a patch went into the stock 2.6.13-rc3 kernel:  
  http://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.13-rc3/2.6.13-rc3-mm2/announce.txt  
  http://www.ussg.iu.edu/hypermail/linux/kernel/0507.2/0039.html  

Patching the kernel for the SyncRAID SR3000/5000 is not easy because it
modifies the _core_ ATA logic.  It's not like other RAID cards that have
their own drivers (via the generic SCSI interface).  But the good news
is that once the drivers _are_ in the stock ATA code, they will be
perpetual.

That's why I bought my SR5000, because once Linux support is there, I
won't have to worry about the ATA driver being "outdated."

> It does _not_ use the BIOS.  NT and Linux kernels can_not_ use the 16-
> bit (Real86) Int13h BIOS services once their 32-bit/64-bit
> (Protected386/PAE52 mode) kernels load.

I didn't mean to be "abrasive" there, but understand there is a huge
difference between 16-bit (Real86) Int13h BIOS disk services and a 32-
bit/64-bit (Protecte386-PAE36/PAE52) kernel driver.

"FRAID" cards rely 100% on the driver for RAID once a 32-bit/64-bit OS
boots, while the OS directly addresses the IDE(s) on each ATA channel --
aka "host-based/software RAID".  True, "Intelligent" RAID cards only
need the 32-bit/64-bit OS driver to talk to its intelligence, and
typically do so through the queuing SCSI subsystem of the OS.

The NetCell SyncRAID SR3000/5000 uses a middle ground.  It uses an on-
board intelligence and even 128MB of DRAM, but it appears as a single
ATA channel with a IDE device.  That way it can work with general ATA
drivers in any kernel (to a point).

> This is many disadvantages, and clearly regulates the approach to more
> desktop environments.  But in that role, the NetCell does excel
> extremely well.

RAID-3 is striping with dedicated parity where there is no blocking.
All disks are written to simultaneously, and then the parity calculated
and committed to its dedicated disk.  During reads, the disks are read
from directly.

This is similar in RAID-4 -- stripes plus dedicated parity -- except
RAID-4 uses blocking for its stripes.  Typical block size is 512 bytes *
65,536 sectors = 32KiB.  You'll note that 65,536 sectors is typical 256
(0-255) heads * 64 (0-63) sectors/cylinder disk geometry = 16,384
sectors, which 65,536 is a multiple of (and an exact multiple if the
media uses 2,048 byte sectors, like optical does).

RAID-5 is blocked stripes but with striped parity.

RAID-3 is ideal for single user applications where maximum, sequential
throughput is attainable.

RAID-4 is better for multiuser applications where many, independent
operations could be going on -- especially large writes.

RAID-5 is better when lots of random access occurs -- especially smaller
writes.

Since NetCell RAID-XL takes RAID-3 and applies it to a specific number
of disks, 3 for 32-bit (2 x 16-bit ATA + 1 parity) and 5 for 64-bit (4 x
16-bit ATA + 1 parity), this is clearly a desktop-solution.  But in
being so, offering a virtual ATA interface is not much of a disadvantage
-- if not an advantage in many applications because of how most OS ATA
drivers work.

For servers, you want to avoid RAID-3 and use blocking stripes, and most
OS SCSI drivers will be most accommodating for such queued operations.


-- 
Bryan J. Smith   b.j.smith at ieee.org      http://thebs413.blogspot.com
--------------------------------------------------------------------- 
It is mathematically impossible for someone who makes more than you
to be anything but richer than you.  Any tax rate that penalizes them
will also penalize you similarly (to those below you, and then below
them).  This is why someone who makes more than you always gets at
least the same, if not a bigger, tax cut.  Otherwise is impossible.