> You can buy some SCSI and IDE controllers whose only purpose in life
> is to look like a single drive to Windows, but actually have
> two identical disks connected, mirroring to the second drive.

A.k.a "RAID1". I'm familiar with the concept; I've been running Adaptec and
Mylex RAID adapters since 1995. I run RAID0, RAID1, and RAID 5 arrays. In
all cases, the OS sees the volume as a single disk drive, though it is
actually composed of 2, 2, or 8 physical drives.

> That's why I mentioned that I didn't know if anyone had such non-RAID
> mirroring hardware for the 3000.  BTW, that's not RAID, just like
> sticking 4 disks on an IDE controller also isn't RAID.

Yes, it (mirroring with hardware, unbeknownst to the OS) is; it's called
"RAID1"; its purpose is to provide redundancy such that a failure of a
physical drive in the array will not result in a loss of data. 4 IDE disks
is probably (OK, definitely) not a RAID1 configuration; I'm not sure how
that got into the conversation. There are ATA-based RAID controllers, but
they do the same thing as their SCSI counterparts: mirror or stripe (or
both) multiple drives, and present the result to the OS as a single device.

> Mirroring, then, has at least five fundamentally different
> implementations:
>    software (OS), on same machine
>    software (OS), with the mirror on a different computer
>    hardware, dedicated controller
>    hardware, modified disk drive, stealthily
>              watching for writes, and ignoring reads
>    hardware, with RAID

Nope. RAID1 is RAID1 is RAID1; whether the mirroring is provided by OS-level
software, by a dedicated intermediate hardware device, by firmware on the
drive, or by some combination of the above, if it manifests a Redundant
Array of Independent (formerly "Inexpensive") Disks, it's RAID. Mirroring to
a different computer is not RAID, because it is not a disk array. All other
members of your list are RAID1, with no "fundamental differences" separating
them.

Of course, we could always ask people who've been doing this for a while.
Among many others:
 http://www.utexas.edu/cc/vms/about/raid.html : RAID1 writes data to two
drives simultaneously. If one drive fails, data can still be retrieved from
the other member of the RAID set. This process is also called
"mirroring."...RAID arrays may be controlled either by dedicated hardware
(controller-based RAID) or by software (host-based RAID).

Alternatively, check the publications of the RAID Advisory Board (RAB), or
go back to the original Berkeley papers by Patterson, Gibson and Katz in '88
and '89: the defining characteristic of RAID1 is, and always has been,
mirroring of data to a second directly-accessible physical drive; whether
that mirroring is achieved by software, firmware or hardware, in the OS, on
the drive(s), or somewhere in between, is irrelevant.

> Note that the key difference between the mechanisms is
> where the "smarts" are:
>    - in the main computer
>    - in the controller (RAID or otherwise)
>    - in the slave (mirror) drive

As noted, where the "smarts" are is specifically *not* a discriminator.

FWIW, the RAB now deprecates the numbered levels for identifying RAID
configurations, due to erroneous perceptions (some vendor-induced) that
higher levels of RAID are "better" than lower.


Steve