[K12OSN] new server serial ATA or SCSI

[Ken Meyer]

My understanding has been that early IDE controllers could deal with only
one of the two potentially connected devices at a time.  That was why it was
recommended that you put your hard drive and CD on different controllers.
Then, at some point, it seemed that the devices could be accessed
independently.  It has never been clear to me how you can tell if they are
or aren't, or even if it is done in the controller or the drivers, but I
have presumed that it is SOP, probably in hardware, by this time.

How many SATA drives can you put on a controller?  SCSI allows a lot, but at
some point, doesn't the PCI bus become the choker?  I recall those big
buffer boards that didn't seem to do much for performance.

Stand-by for new 3Ware boards.  Don't know whether they will leverage
on-drive TCQ, or will do it themselves though.

Ken Meyer


----- Original Message -----

From: Quentin Hartman <qhartman at lane.k12.or.us>
To: <k12osn at redhat.com>
Sent: Wednesday, February 18, 2004 11:18 AM

Subject: RE: [K12OSN] new server serial ATA or SCSI


> On Wed, 2004-02-18 at 08:34, Les Mikesell wrote:

> > I haven't tested these myself but I think they still only accept a
single
> > command at a time.
>
> The new 73GB Raptors (but not the 36GB ones!) support the tagged command
> queueing, so they are very SCSI-like in that respect. I've used these
> drives in one of my client's machines and recommend them _highly_. They
> are _fast_. And with a standard 5-year warranty, they are hard to beat.
>
> <snip>
>
> >  The big advantage of scsi is the ability of the controller to
> > multiplex seeks across several drives at once.
>
> <snip>
>
> I'm not sure that I am addressing specifically what you are talking
> about here, but it sounds like you are referring to ATA devices only
> being able to access one device on a chain at a time, correct? In SATA,
> this problem does not exist, each device is on a separate channel and
> all can be accessed simultaneously. With the lower-end (read: onboard /
> promise) chipsets, there is pretty significant CPU overhead though.
> Better controllers from 3Ware are supposedly much better about this, but
> I haven't seen any real benchmarks or had any hands on experience with
> them, so it could be purely hearsay.
>
> --
>    -Regards-
>
> -Quentin Hartman-
>
> Academic Computing and Networking Services Coordinator
> Fern Ridge School District 28J
> Elmira, OR
> Office: 541-935-2253 x429
> Cell: 541-914-2989
> qhartman at lane.k12.or.us
> www.fernridge.k12.or.us