dtrace discuss - Jan 2010 - Concerning FC/"SAN" capability of FISHworks/7000 series appliances.

Hi all.

I''m just looking for some confirmation on functionality of stuff I
can''t seem to get my hands on, as yet.

Concerning the 7000 series boxes - I note the "SAN" tab in the updated
version of OpenSolaris ''firmware'' running on the appliances,
and I''m aware that the FC stack stuff was at least, at the end of last
year ''any day now''.

I just wanted to know how the 7000 series kit presents storage over FC or
"SAN", as the tab is known.

I.e - will it *only* present pooled storage (zvols/zpools etc), or can it
present raw physical media at the block level. I.e - can I have an SSD module
sitting in a 7410, in a J4400 connected below, and select this device
exclusively to be presented on the FC loop, excluding the utilisation of any
L2ARC or ZFS etc? I''d like to be able to present disk just as disk, and
not ''baked'' through ZFS, for the sake of argument...

I''m assuming the answer might be no, on the basis of this not being the
whole ethos of the appliance, but since I haven''t got any concrete yes
or no answers so far, maybe here is the best place to ask!

Thank you, all.

z
-- 
This message posted from opensolaris.org

> I''m just looking for some confirmation on functionality of stuff I
can''t seem to get my hands on, as yet.
> 
> Concerning the 7000 series boxes - I note the "SAN" tab in the
updated version of OpenSolaris ''firmware'' running on the
appliances, and I''m aware that the FC stack stuff was at least, at the
end of last year ''any day now''.
> 
> I just wanted to know how the 7000 series kit presents storage over FC or
"SAN", as the tab is known.
> 
> I.e - will it *only* present pooled storage (zvols/zpools etc), or can it
present raw physical media at the block level. I.e - can I have an SSD module
sitting in a 7410, in a J4400 connected below, and select this device
exclusively to be presented on the FC loop, excluding the utilisation of any
L2ARC or ZFS etc? I''d like to be able to present disk just as disk, and
not ''baked'' through ZFS, for the sake of argument...
> 
> I''m assuming the answer might be no, on the basis of this not
being the whole ethos of the appliance, but since I haven''t got any
concrete yes or no answers so far, maybe here is the best place to ask!
You definitely get points for creativity. ;)  And at the risk of overly
rewarding it:  no, you cannot directly export raw spindles via the block
interface of the 7000 Series; it must be backed by ZFS.  Two notes on
that:  first, we allow you to tune ARC utilization on a per-LUN basis,
so you can (for example) enable the L2ARC on only for the LUNs that you
want to use the L2ARC.  Second, note that we support multiple pools in
our forthcoming release, which allows you to have different LUNs that
have different underlying storage properties.

Finally, in terms of schedule:  we''re wrapping up the release that
contains
FC currently, and it''s available for your immediate evaluation (with
some
modest provisions).  If you are so inclined, mail me privately.

	- Bryan

--------------------------------------------------------------------------
Bryan Cantrill, Sun Microsystems Fishworks.       http://blogs.sun.com/bmc

Bryan,

Thanks for the response! Appreciated.

As I expected, really. 

I''m trying to solve a problem with a large archiving solution at the
moment and attempting to see if I could leverage the 7000 series as the
centre-part of the equation. The premise of attaining great performance through
SSD for meta slices was enticing.

Architecturally, I don''t think it is the right ''fit''
for what I''m attempting (shared SAM/Q pools, with meta data slices on
the FC loop sitting on SSD - with underlying cache disk controlled via ZFS for
QFS clients and meta controller to see on multiwriter/multiread configuration).
This doesn''t work, as a result of ZFS''s multi mount/non
concurrent mount limitations. I.e - you could never have the filesystem mounted
in more than one location, notionally, at any one time. Currently, a
multi-reader, multi-writer, it is not!

I guess it brings me to another question really. Is there, or are there any
plans or considerations to one day somehow bring a multi-reader/multi-writer
model to ZFS? Maybe it goes against design principles to begin with, and as
such, isn''t really a sensible possibility?

Thanks anyway!

z
-- 
This message posted from opensolaris.org

> I guess it brings me to another question really. Is there, or are there any
plans or considerations to one day somehow bring a multi-reader/multi-writer
model to ZFS? Maybe it goes against design principles to begin with, and as
such, isn''t really a sensible possibility?
That''s probably more appropriate for the ZFS discussion list, but, no,
there is no plan for multiple active writers for example.

Adam

--
Adam Leventhal, Fishworks                        http://blogs.sun.com/ahl