Dear group,

I have a question. I have a Oracle 10.2.0.3 database, everything is
on SAN storage ( Raid 10). When i am asking for another LUN for 2nd
redolog members, he told me we have already mirrored for u at the san
level. So he thinks it is not necessary to have two members of
redolog in each group.

Could somebody share some idea with me?

Thanks

On Apr 29, 2:33*pm, charles <dshprope... (AT) gmail (DOT) com> wrote:
This is not exactly a new type question. Various people will have
different answers here.

On most of my dev/test databases ... no I don't use it multiple
members in a redolog group usually.

Production databases I will go case by case. Highly critical systems
sure why not.

Have I ever lost a redolog group or just one member of a group? Not
so far ...

Here is our SA's comment

I'm NOT fine with this but please remember that I HIGHLY recommend
against this. It is a best practice to never do software raid which
is what you are basically doing. This is very old practice when you
don't have the infrastructure that we have.

Charles:

# Here is our SA's comment ...

I'm NOT fine with this but please remember that I HIGHLY recommend
against this. *It is a best practice to never do software raid which
is what you are basically doing. This is very old practice when you
don't have the infrastructure that we have.

....

Kind of doubt that your SA has seen as many failure scenario's as
Oracle customers have experienced and Oracle coders have engineered
against.

You can practice on a test system by having an active redo log group
with multiple members. Trash one of the current members ( on a linux
system dd command can be used ) while Oracle is using both members.

Destructive testing of failure scenarios on a test system is a good
thing.

Even if you don't have any hardware failures there is always a
possibility of human error.

On Apr 29, 3:12*pm, charles <dshprope... (AT) gmail (DOT) com> wrote:


Ask him if he would bet his job on that quote. Actually in Oracle, if
you do not do "software RAID" (ie multiple redo log members/groups) he
needs to have his head examined. I don't know where he pulled that
"best practices" quote from, but it was not from anything I have seen
in the Oracle world going back 20+ years. To "bet" that your hardware
RAID will save your bacon is extremely confident in technology that
proves itself over and over again that if it can fail, it will and is
not a matter of "will it fail" but "when will it fail" AND usually at
the most inconvenient time.

If he takes that view of things, then I guess that he would never use
ASM (which, in the *n*x world, would be a huge mistake).

Not sure how big your environment is, but if he really wants to play
that game, tell him to giver you {N} equally sized LUNS, install ASM
and tell him to get out of the way. Those LUNs can be on RAID10 or
even RAID5 or 6. and doing so, may just increase your performance.
Then you can manage your space instead of an uninformed SA. If the
database is mission critical, you may want to also look at ASM FAILURE
groups. Cool stuff and runs really fast.

On 29 apr, 21:12, charles <dshprope... (AT) gmail (DOT) com> wrote:
Hardware mirroring will not protect you against redo log corruption,
the corrupted log will just be happily mirrored.
Have your SA read up on what Oracle calls a "redundancy set"; a good
start would be:
http://www.google.com/search?q=oracl...set%22&tbm=bks

HTH

--
Jeroen

On Apr 29, 3:12*pm, charles <dshprope... (AT) gmail (DOT) com> wrote:
Tell your SA that the Raid-10 will protect the online redo log from
disk media failure but it will not protect the log file from logical
corruption which is the point of having multiple online redo log
members in a redo log group. If you check your DBA Administration
manual I believe you will find Oracle recommeds having mirrowing the
online redo logs at the Oracle level, that is, two or more redo log
file members per group.

Can you live with only one, yes. The odds are in your favor that you
will never have an outage due to corruption of the logs, but if you
have multiple redo log members the odds are even better. I would make
the decision based on available disk space and performance impact.

HTH -- Mark D Powell --

On Apr 30, 4:33*am, charles <dshprope... (AT) gmail (DOT) com> wrote:

Been discussed many times. Recently at the oracle-l list the
consensus even from the Oakies is do not do mirroring twice. Either
do Oracle multiple members, or do SAN-based mirroring. Doing it twice
achieves really nothing other than slowing things down. One or the
other, not both.

Given that the days SANs lost data are long gone (not a single data
failure in over 4 years now of 12TB of I/O per day) we've gone for SAN-
based last year. Fine so far and archived redo logs copied to DR site
every two hours gives me even more comfort.

YMMV based on how paranoid you are.

On May 2, 1:20*am, Noons <wizofo... (AT) gmail (DOT) com> wrote:
Would you really come to the same conclusion without DR? In my view,
applying the archived logs to a standby is the only way to know the
redo stream is at least non-corrupt, and if you don't have that, you
want to have additional checks elsewhere.

You may be over-estimating everyone else's ability because your ship
is so tight! I'm not running standby now because a particular array
works with two versions of hp-ux, but not the version in between, and
no one knows why. SAN's are not magic, they fail when their
components get old or their environment changes. The failure may lack
grace.

jg
--
@home.com is bogus.
Not paranoid enough: http://storagemojo.com/2011/04/29/amazons-ebs-outage/

Joel:

#*SAN's are not magic, they fail when their components get old or
their environment changes. *The failure may lack grace.

SAN's are built up from storage arrays and every component in a
storage array can fail and often in multiple diferent ways. Redundant
components are included typically but every different component has a
set of inherent weaknesses.

Not to miss mentioning software problems/bug and human error/mistakes/
deliberate hacks.