"John Rumm" wrote in message
o.uk...
On 24/06/2010 09:08, dennis@home wrote:


"John Rumm" wrote in message
o.uk...

Some early implementations of RAID level 2 tried this. Bit level
splitting of data over some drives and applying FEC to generate check
and correct bits on parity drives.

Its not used these days since for obvious reasons - and it kind of
went against the whole raid philosophy in the first place, so calling
it "proper RAID" is a Dennis'ism really.

Show me where I called it a proper RAID?

Erm , how about a couple of posts back where you said "Ah well that was
probably in the days of proper RAIDs. The ones where it was done bitwise
across the disks and all the spindles and heads were synchronised.
They were expensive."

You are reading stuff that hasn't been written.

See above...

PS what do you think are the obvious reasons they died out?

"Died out" is perhaps over egging it - it never really got started. There
were several problems; Price was a significant factor in the first place -
it required bespoke drives

Well that's not exactly true, most of the drives at the time had sync
connectors and you just didn't use them if you didn't need them.

and controllers with non standard interfaces.

The interfaces were standards at the time there was nothing special about
the drives compared to other drives.

It also used a comparatively large number of drives compared to other RAID
setups as well, without providing the performance or redundancy advantages
either.

They were certainly redundant.
Performance relative to single drives was very quick.

Most drives rotated quite slowly and it took a lot longer to read the data
from a single drive than an array, the latency was the same.


As a technology it was rendered obsolete almost immediately when the drive
manufacturers included at first equal (and shortly later, superior) FEC
within their drive firmware.

There was little or no firmware on drives at the time.
The integration of controllers did kill them off.
People like me were responsible as we decided to use SCSI and make the
controller designers do something more useful.
At the time every computer would have its own controller design, that was a
complete waste of time when there weren't enough engineers around to design
more useful bits like bit slice CPUs and bubble memory cards!

I actually went to a disk drive conference (I don't really know why) in the
early eighties, there were some real die hard engineers there that would
come along and sensing that I didn't design controllers start spewing
un-decipherable jargon. They looked a bit shocked when I said we were going
to use SCSI as you couldn't even buy a SCSI disk at the time. A year later
things were different.

That meant that a pair of mirrored drives on standard controllers offered
better reliability at a fraction of the cost. So game over for RAID 2.

I never saw a mirrored pair of drives at the time, it just wasn't going to
happen as you needed the parity bits to do correction and redundancy.