A1000 SCSI questions

Fri, 14 Jan 2000 05:43:39 +1000 (EST)

Err, I'm somewhat without sleep, lets see if I can follow you...

You could use the h/w mirror *within* the array, then stripe
bunches of them (within and/or between the arrays), but this
isn't good as there's a single controller in the A1000, and
you don't want to be dependent on it if you've got >1 array
(for less critical workgroup servers non-mirrored controllers
is probably okay, though I'm still not keen on the idea).

Easiest is probably to define each disk as a LUN (up to 12 LUNs,
which is within the 16 LUN limit with appropriate patches),
then initialise them within VxVM and do what you like. This
will still get the benefit of the controller's cache, which
is the main benefit anyway (you wouldn't do this for RAID-5
though, for that you'd use the controller to get advantage
of it's optimisation algorithms--eg. it knows it doesn't
need a disk-based RAID-5 transaction log, it can just do
it into battery-backed memory, plus the reads of existing
data and/or parity for R5 updates can be hidden from the
application once it's sent the new data block to the array).

Alternatively, you could define stripes within the A1000,
then mirror them with DiskSuite/VxVM--however there is the
full recovery time to consider, plus hot spares become
problematic--the A1000 can't do it as it doesn't have
access to the other A1000, and you need to define a really
big hot spare for SDS/VxVM to use, because if one disk in
the stripe fails it thinks the whole disk fails (looks like
a disk with very flakey media!)

Pairs of A1000 are still a good way of getting relatively
cheap mirrors with write caching in the storage (eg. for
use with clusters, where the cache must be accessable to
another node in the event of failure). A3500(FC) is better
yet, but much more expensive for small capacities.

HTH,

Stuart.

ps.  Usual disclaimer re my opinions as opposed to Sun's...

`  > You can use DiskSuite or SEVM/VxVM on top of the A1000 on-board RAID,
`  > but this doesn't solve the single bus issue.
` 
` Ah, a second interesting question is coming to my mind. The VM manual
` says that striping above mirroring helps to improve recovery time
` after a disk fault (the stripes need to be large enough so more than
` one subdisk is being used in one stripe, in effect the subdisks
` themselves will be mirrored, thus being recovered instead of the
` complete mirror).
` 
` But it seems to me that this concept cannot be implemented using the
` A1000 hardware raid feature (rm6). It has to be done by using VM
` exclusively which means that the valuable Cache Memory of the A1000
` arrays cannot be used. Is this correct?
` 
` Andreas

---------------------------------------------------------------------
 ,-_|\  Sun Microsystems Australia  Stuart Remphrey, Systems Engineer
/     # P.O. Box 86, Albert St    stuart.remphrey@Aus.Sun.COM x.56308
\_,-\_/ Level 10, 80 Albert St   Ph (07) 3238 8308 Fax (07) 3238 8320
     v  Brisbane  QLD  4002       +61 7  3238 8308   +61 7  3238 8320
---------------------------------------------------------------------
Order a StarOffice CD in Australia: http://www.sun.com.au/staroffice/