A3000 & Ultra II questions (long!)

[Brian Herzog]

> [...]
> Some of our concerns are:
> 
> 1. What happens when a cache battery fails and needs to be replaced?

Fast writes are automatically disabled, and notification sent that the
battery needs to be replaced.

> Can you do this while the system is live?

Yes.

>  Does the system fail
> gracefully to a degraded level while still working transparently? 

Yes.  Writes must go all the way to the disk before they are
acknowledged to the host.  This eliminates any possibility of data loss
should power fail with writes in cache but not yet committed to disk.
Otherwise, it's business as usual.  The system keeps on running and I/O
continues to occur.

> Or
> does it lock down write access to the disks until you can replace/fix
> the battery?

I'm afraid I don't know what you mean by "lock down write access to the
disks".  Please clarify.

> 2. Same question(s) for disks and controllers.  In the heat of the
> moment, how picky is the system about pulling a bad disk and replacing
> it with a new one?  Is there alot of work to be done?  Do you need to
> bring down the host system at all to make changes?

The work to be done is to hot-swap the replacment disk.  Everything is
handled automatically by the A3000.  There is no reason to bring down
the host, or even interrupt the application(s), for that matter.

> 3. Since ODS seems to be going away in favor of the Veritas Volume
> Manager, should we also think about getting the Veritas VxFS stuff as
> well?  I'm honestly interested in being able to grow *and* shrink
> filesystems as needed.  I'm thinking that using the A3000 to make one
> huge RAID5 set and layering multiple filesystems on top of it would be
> the way to go.  And if I can setup file system quotas, then that gives
> more flexibility as well.   Time to read the white papers more closely.

Of course, A3000, being a hardware RAID device, has its own volume
management (called RAID Manager), and presents logical volumes to the
host.  That said, there are circumstances and reasons why you would
want to use a host-based volume manager on top of that, and you are
free to do so.  Note that Veritas Volume Manager (VxVM) has been tested
and qualified for use on top of A3000; SDS has not.

File system is an independent subject from volume management.  A3000
just presents logical volumes to the host; you can do whatever you want
with them.  Raw, UFS, VxFS, whatever.

> 4. How many disks min/max can I have in a RAID5 set on the A3000?
> Does the Disk Manager take care of this?

RAID Manager will automatically take care of this, based on the size of
volume you're trying to create.  Default is 6 disks (5+1) because there
are 6 back-end buses. (If the RAID set is smaller than this, then RAID
Manager will take a slice out of the 6 disks).  You can force it to use
fewer disks; I suppose the minimum is 2, although I cannot think of any
reason to use less than the default.  Maximum is 20.

> 4. How easy is it to grow a RAID5 setup with more disks?  Does the
> RaidManager 6.1 (RM6.1) software allow you to do this?

Not at this time.

>  Or do we need
> to build seperate RAID5 sets and stripe them together? 

Yes, or create a separate set of the desired size and move the data
over to it.

> I'm also
> looking at a Symbios RAID controller and it seems to allow this to be
> done.

It does not.

>  And since the A3000 is based on their controller, I would hope
> it would be possible.

Your logic is fine; if the Symbios controller did, so would the A3000.
But it doesn't.

> 5. How does the A3000 handle hot spares and the rebuilding of RAID
> sets?  When you replace the failed disk, does the system rebuild back
> onto the new disk and then return the global hot spare disk to the hot
> spare pool?

Yes.  Automatically.

> 6. How hot does a fully loaded system with 35 x 9.1Gb disks run?  I'm
> starting to run into cooling issues in my machine room and I'd like to
> be a little more pro-active if I can.  

It runs within the specs on the A3000 data sheet. (Sorry, I don't have
the numbers in front of me.)

-Brian