Adaptec AAA-13x RAID

[mike at bilow.com (Michael Bilow)]

You do not really get much with such "hardware" RAID controllers under
Linux.  As far as I know, all of the low-priced ones actually do software
RAID, and they come with drivers to provide software RAID under Windows or
whatever.  While they work with Linux, their primary value is in providing
additional physical IDE channels at non-conflicting addresses.  Since only
one drive per IDE channel can be active at thr same time, putting two
drives which participate in the same RAID set onto the same physical IDE
channel would introduce a substantial performance penalty.

This means that, with a conventional motherboard that has only two
physical IDE channels, RAID-1 (full mirroring) is the best possible
approach.  If you want to do RAID-5, which makes much more efficient use
of storage, then you really need to have either three or five independent
physical IDE channels, one for each drive.

Two 10GB drives in a RAID-1 configuration would provide only 10GB of
useful storage, a 50% penalty.  RAID-5, on the other hand, is typically
implemented either with three devices or five devices.  Using RAID-5,
three 10GB drives would provide 20GB of useful storage, a 33% penalty, and
five 10GB drives would provide 40GB of useful storage, a 20% penalty.

In order to provide enough physical IDE channels for efficient software
RAID under Linux, these specialized "hardware" RAID controllers can be
very helpful.  Other than that, I am not sure they are worthwhile.  You
can never get IDE devices to hot-swap, for example.

None of this reasoning applies to SCSI, it is important to note.  With
SCSI, there is essentially no performance bottleneck which results from
connecting multiple devices, even those which participate in the same RAID
set, using the same SCSI bus.  In extreme cases, there are certainly
issues with doing this, but in practice the concern is not significant
because the speed of a modern SCSI bus is several times faster than the
sustained data rate of any disk drive.  SCSI, at least with devices that
all support disconnection, is quite efficient at sharing bus bandwidth.

-- Mike


On 2000-12-13 at 20:10 -0500, Randall Hofland wrote:

> I thought it worth noting that several modest hardware RAID options for IDE are
> available and a more advanced unit is in the works. ABIT makes an inexpensive 2
> channel/4 device 0+1 RAID PCI card using the Highpoint controller chipset
> (about $40-$50) and they have LINUX drivers plus their Gentus LINUX, Promise
> makes a more expensive but otherwise similar PCI card and there is a new unit
> coming out with 4 channel/8 device that sounds really hot but I haven't seen it
> yet on Pricewatch. Abit and other also have the ATA/100-RAID options built into
> some of their boards: I highly recommend the Abit KT7-RAID as a great low cost
> option.
> 
> Michael Bilow wrote:
> 
> > I was actively involved with this and many of those messages might have
> > been written by me.  There is not and never will be 7810 support in Linux
> > while Adaptec provides documentation for it only under non-disclosure.
> >
> > Note that new-style software RAID in the kernel is quite solid.  You would
> > not get hot-swap, but the basic advantages of RAID are there.  Considering
> > the prices of large IDE drives these days, using software RAID over a bank
> > of IDE drives is worth considering.  At today's prices, $500 would buy you
> > two Maxtor Diamondmax 80GB IDE drives which could be mounted on the two
> > channels of a standard motherboard for RAID-1.
> >
> > -- Mike
> >
> > On 2000-12-07 at 12:34 -0500, Matthew J. Brodeur wrote:
> >
> > >    Does anyone know if there is or ever will be support for the Adaptec
> > > AAA series RAID controllers under Linux?  The one I have is the AAA-131SA,
> > > UW single channel SCSI.  The aic7xxx driver recognizes the SCSI channel
> > > (7880), but produces an unhappy message about not supporting  the array
> > > controller (7810).
> > >    I did some searching, but all I found were old messages indicating that
> > > no work was being done on a driver. Is this still the case?

-
Subcription/unsubscription/info requests: send e-mail with
"subscribe", "unsubscribe", or "info" on the first line of the
message body to discuss-request at blu.org (Subject line is ignored).