Why would anybody want to use any other redundancy than 'external'? The
'normal' redundancy means that there is 2-way mirroring and 'high'
redundancy means that there is 3-way mirroring. My servers are connected
to SAN (HP EVA 8000) which does RAID 1+0, RAID-5 or RAID-6, as needed.
There is also multipathing enabled, which means that there are 2 HBA's
and that if one fails, the other one takes over. Significant investment
was made into SAN with all those capabilities, 16G of NVRAM, storage
virtualization, snapshots and many other goodies. A consultant is now
telling me that I am doing things wrong because I settled for an external
redundancy. Also, I keep voting file and OCR on an OCFS2 file system,
precisely to avoid the local registry stuff. I am aware of the
recommendation to keep everything on ASM but that complicates the install.
Oracle licenses are expensive as it is, I have a limited number of CPU
cores at my disposal and I have to make the best of it. Why would I need
to waste CPU on ASM mirroring when the company has already paid for the
device that can do it automatically, with its own CPU?



--
http://mgogala.byethost5.com

On 7 Sep., 15:28, Mladen Gogala <gogala.mla... (AT) gmail (DOT) com> wrote:
Hi,

we use the mirroring feature delivered by ASM at two sites.
And we do this most of all because of saving money

Think about a two node RAC, each node sitting in a separate data
center at
the same building, but in different rooms because of security reasons.
Fire or water
will destroy normally only one site (hopefully)

So, regarding this scenario, you will normally use two storages, one
for each data center.

Now you have the choice, use the ASM mirroring feature to make all
mirroring over both
storages or use some build in feature delivered by the storage
manufacturer. Last one will
often cause expensive investments, because this feature is not cost
free.

In this case we use ASM mirroring with two failure group, one on each
storage. So we have
no single point of failure.

Bye

Joerg

On Sun, 18 Sep 2011 03:38:56 -0700, Jörg Jost wrote:

That's the best motive I know of. IT as such was introduced to save money
in the first place. Before the IT, companies were working like in the
opening scene of Monty Python's "Meaning of Life": an army of
accountants, all with mechanical calculators, supervised by a "slave
master". Now, 3 Dell boxes, clustered over NetApp are much cheaper than
the army of accountants, even with Oracle RAC.



Now, this is a perfectly reasonable explanation: ASM does for free what
the storage manufacturer charges for: replication between 2 different SAN
devices. Makes perfect sense. How are you satisfied with the performance?
That, of course, means that the server CPU cycles are used for disk
mirroring. Is it a big burden? The companies that I worked for have
usually used external redundancy, provided by the SAN itself plus a
standby database, usually in maximum performance mode, to minimize the
impact on the primary. In one case, the standby replication went over WAN
from Louisville, KY to NYC, NY. It did consume very significant
communication resources, completely devouring a T2 line.


--
http://mgogala.byethost5.com

On Sun, 18 Sep 2011 17:48:07 +0000, Mladen Gogala wrote:

In case you haven't seen it, look for the "Crimson Permanent Assurance"
on Youtube.



--
http://mgogala.byethost5.com

On 18 Sep., 19:48, Mladen Gogala <gogala.mla... (AT) gmail (DOT) com> wrote:
The performance of the disc subsystem makes no trouble at all.
But i can not compare the performance to the alternative solution
delivered
by the storage build in feature because of the mentioned reasons.

Oracle 11g is at least theoretical better in doing this than 10g. The
feature called
preferred read failure group is very helpful for a configuration
described above.
We have both versions in production, but the sides are so different
that i can
not compare the numbers without comparing apples with pears.

And at Monty Python, i love them, i own the mentioned film on DVD and
some
more, so thanks for recalling this, i have to look this film again

Bye

Joerg

Jörg Jost wrote,on my timestamp of 18/09/2011 8:38 PM:

Hmmm..... Weird... I've never seen anyone save money when using ASM...


And ASM does the mirroring out of what? Thin air? Doesn't it need a high speed
interconnect, just like the storage replication does? It's not included in the
cost of Oracle...
Sorry, but I don't see why ASM has to be cheaper when it needs the same high
performance hardware connection as the storage.
And please, don't anyone jump in and tell me that ASM is "highly optimized": it
isn't, and that is NOT what I am talking about.

On 19 Sep., 12:16, Noons <wizofo... (AT) yahoo (DOT) com.au> wrote:
As i said, the feature to mirroring data over two storage devices is
not for free. At
least not by every manufacturer. Of course inside one device,
mirroring is build in.
And of course you have to connect the two devices over fast
connections.

All what this discussion is about is mirroring the writing of the
database server
processes to another SAN. Oracle writes only to one target, but for
security reasons
you need the changed blocks on another SAN located anywhere.

So the only cost difference lies in the money you have to spend for
the mirroring feature
to the storage manufacturer. Costs for hardware including network
devices is the same
for both solutions.

Bye

Joerg