RE: Was (RAC ASM) is ASM + NAS
- From: "Matthew Zito" <mzito@xxxxxxxxxxx>
- To: <kevinc@xxxxxxxxxxxxx>, "Oracle-L Freelists" <oracle-l@xxxxxxxxxxxxx>
- Date: Mon, 12 Jun 2006 16:03:02 -0400
ASM-over-NFS definitely strikes me as silly - I certainly understand the
appeal of SE, but the requirement for ASM is downright obnoxious.
Obviously, let's say you had a Netapp with only NFS licenses and wanted
to roll out your first SE RAC cluster, sure, you're between a rock and a
hard place. However, with Netapp giving away free iSCSI licenses,
that's less likely. Same-tier abstraction layers on top of abstraction
layers are rarely elegant solutions.
With regards to the performance, I see where you're heading (aka why not
use a commodity cluster filesystem-driven solution where you can scale
in individual "data mover" elements). To answer your question, though,
I've definitely seen filers get saturated - for example, our 270cs here
in the office can very easily get saturated in heavy mixed-workload
environments. However, I was just futzing with a 3020 non-clustered,
and saturated a single gigabit link on the storage side with about
25%-50% overall CPU utilization and low response times (I wasn't
intentionally trying to benchmark the storage, just prove the
environment was set up correctly, so no pretty graphs or exact stats).
If we really wanted high performance NAS hardware of the non-commodity
sort, the BlueArc titan offers ridiculously high throughput for a
correspondingly high price tag.
I'm a big believer in commodity everything - I'm a big proponent of
iSCSI (picked a lot of fights with folks at EMC while I was there about
that), so I'm a definite believer in the HP storage gateway model (a
number of polyserve data movers exporting data via NFS with HP storage
on the backend). What I like about Netapp is the universal-ness of the
product (for lack of a better/real word). So - not just NFS, but CIFS,
Fibre Channel, iSCSI all in the same platform. I can upgrade the
product by ugprading the head or controllers. The FlexVol architecture
is a dream in terms of managing complex environments, and the read-write
snapshots in FlexClone can be used for a whole slew of different pieces
of functionality. To quote our senior development manager when we
upgraded our first filer to Data OnTap 7 and turned on flexvols, "Wait,
that's it? It's that easy? Why did we wait 6 months to upgrade?"
So, for sure, you can scale cost-effectively to very very good
dollar-per-nfs ops ratio using commodity solutions, and I think in pure
speeds-and-feeds environments that's a great solution. What I think is
a lot more interesting is the general-purpose case, where people need
storage that is easy to use, flexible and dynamic, reasonably good
performance, and scalable in capacity.
Thanks,
Matt
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Kevin Closson
Sent: Monday, June 12, 2006 3:22 PM
To: Oracle-L Freelists
Subject: Was (RAC ASM) is ASM + NAS
>>>I actually really like the NetApp model - its very easy to manage,
very flexible, and the multi-protocol piece works
>>>great for us. We have several hundred systems (mix of real and
virtual) that use netapp as their storage in our
>>>development lab. With Netapp, we can expose the same data via fibre
channel, iSCSI, and NFS, so we can easily
>>>distribute storage resources among the different environments. And
the WAFL model is one that is
>>>tried-and-true, in that its strengths are well-known, its weaknesses
are pretty-well documented at this point (though
>>>they've been mitigated in OnTap 7). Because of that, you can expect
consistent results across different versions of
>>>OnTAP, and different netapp hardware platforms. Compare that with
HP, HDS, EMC, where different families of products
>>>(of course) have vagaries, but there are even fundamental shifts
between revisions within the same family.
...excellent post, Matt. Espousing technology for technical reasons. Now
THAT is something I repsect!
Do you think my sentiment about ASM on NAS (file mode) is off base? I
can see how the ASM->iSCSI model would make a lot more sense.
Thanks for pointing out the NVRAM point.
So, have you ever had, say, a 4 node RAC cluster saturate a filer? What
did you di to mitigate that condition ?
--
//www.freelists.org/webpage/oracle-l
--
//www.freelists.org/webpage/oracle-l
Other related posts:
