Data Storage - SATA raid: single point of failure?

Author

SATA raid: single point of failure?

Jim Wall

2004-06-10, 12:00 am

With SATA drives in a raid array, is the raid controller always going
to be a single point of failure? In other words, is there any way to
have two fully independent raid controllers accessing the same SATA
drive?

Here is more detail of my concern. Using FC drives, I have two fully
redundant arbitrated loops all the way to the drive. So with redundent
RAID controllers, I have no single point of failure. With SCSI drives,
I can have multiple initiators on the bus, so with redundent RAID
controllers, I again have no single point of failure. What can I do
with SATA drives to avoid this?

Thanks!

Nik Simpson

2004-06-10, 12:00 am

Jim Wall wrote:
> With SATA drives in a raid array, is the raid controller always going
> to be a single point of failure? In other words, is there any way to
> have two fully independent raid controllers accessing the same SATA
> drive?

That's going to be a function how the controller(s) and the backplane is
designed, there is nothing specific to SATA that makes a redundant dual
controller setup impossible. Infact, the rececntly announced EMC/Dell AX
product has a dual controller configuration.

>
> Here is more detail of my concern. Using FC drives, I have two fully
> redundant arbitrated loops all the way to the drive. So with redundent
> RAID controllers, I have no single point of failure. With SCSI drives,
> I can have multiple initiators on the bus, so with redundent RAID
> controllers, I again have no single point of failure. What can I do
> with SATA drives to avoid this?

The only single point of failure I can think of would be the drive
connection to the backplane, and that could fail regardless of whether the
array is SCSI, FC or SATA. Assuming the backplane is designed properly such
a failure should only knock out a single drive.

--
Nik Simpson

Thor Lancelot Simon

2004-06-10, 12:00 am

In article <c2c855a9.0406091409.4ee95425@posting.google.com>,
Jim Wall <jimwall2000@yahoo.com> wrote:
>With SATA drives in a raid array, is the raid controller always going
>to be a single point of failure? In other words, is there any way to
>have two fully independent raid controllers accessing the same SATA
>drive?
>
>Here is more detail of my concern. Using FC drives, I have two fully
>redundant arbitrated loops all the way to the drive. So with redundent
>RAID controllers, I have no single point of failure. With SCSI drives,
>I can have multiple initiators on the bus, so with redundent RAID
>controllers, I again have no single point of failure. What can I do
>with SATA drives to avoid this?

For the difference in price, you should be able to buy two entire SATA
RAID arrays and mirror them.

As to the question of multiple initiators on the same SATA bus, I _think_
the answer is no; but maybe someone more familiar with the SATA spec can
give you a definitive answer.

--
Thor Lancelot Simon tls@rek.tjls.com
But as he knew no bad language, he had called him all the names of common
objects that he could think of, and had screamed: "You lamp! You towel! You
plate!" and so on. --Sigmund Freud

Nik Simpson

2004-06-10, 12:00 am

Thor Lancelot Simon wrote:
> In article <c2c855a9.0406091409.4ee95425@posting.google.com>,
> Jim Wall <jimwall2000@yahoo.com> wrote:
>
> For the difference in price, you should be able to buy two entire SATA
> RAID arrays and mirror them.
>
> As to the question of multiple initiators on the same SATA bus, I
> _think_ the answer is no; but maybe someone more familiar with the
> SATA spec can give you a definitive answer.

I very much doubt that the SATA arrays are implemented in this way, much
likely to use a switched internal architecture with a seperate SATA channel
to each drive, controllers would probably be the same, i.e. each controller
going into a switch port so that it can address all the drives
independently, such an arrangement should allow for failover of all drives
from one controller to other.

--
Nik Simpson

Bill Todd

2004-06-10, 8:06 am

"Jim Wall" <jimwall2000@yahoo.com> wrote in message
news:c2c855a9.0406091409.4ee95425@posting.google.com...
> With SATA drives in a raid array, is the raid controller always going
> to be a single point of failure? In other words, is there any way to
> have two fully independent raid controllers accessing the same SATA
> drive?
>
> Here is more detail of my concern. Using FC drives, I have two fully
> redundant arbitrated loops all the way to the drive. So with redundent
> RAID controllers, I have no single point of failure. With SCSI drives,
> I can have multiple initiators on the bus, so with redundent RAID
> controllers, I again have no single point of failure. What can I do
> with SATA drives to avoid this?
>
> Thanks!

I have a vague recollection that SATA II may support multiple initiators in
some way. Until then, the fact that SAS supports SATA may allow you to use
multiple initiators (depending upon the availability and conformance of SAS
hardware).

Having dual connections right to the individual drive may be over-rated,
since of course the drive itself remains a single point of failure (and
perhaps a more likely one than the busses it's connected to). But having
multiple initiators on a relatively robust bus (which is unlikely to be
taken down by the failure of any single device on it) has actual value,
since it isolates storage failure from server failure (though when the
'server' is simply a controller its failure rate may be sufficiently
eclipsed by that of the devices it controls that controller redundancy can
be largely ignored - you just make sure that your redundant storage uses
separate controllers to its multiple copies or stripe disks).

It should almost go without saying that if you do employ redundant RAID
controllers, they must be able to cooperate such that the back-up is aware
of the relevant state of the array when fail-over occurs.

- bill