Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Usenet -> c.d.o.server -> Re: Unable to manually create RAC database

Re: Unable to manually create RAC database

From: The Boss <usenet_at_No.Spam.Please.invalid>
Date: Tue, 18 Dec 2007 00:03:13 +0100
Message-ID: <47670031$0$85779$>

DA Morgan wrote:

> The Boss wrote:
>> DA Morgan wrote:
>>> The Boss wrote:
>>>> DA Morgan wrote:
>>>>> trub3101 wrote:
>>>>>> On 15 Dec, 12:59, DA Morgan <> wrote:
>>>>>>> trub3101 wrote:

>>>>>>>> Hi Guru's,
>>>>>>>> I am hoping that someone (anyone!) will be able to provide me
>>>>>>>> with some assistance with this as my RAC experience is very
>>>>>>>> limited. I have been trying to manually create a new RAC
>>>>>>>> database <new_RAC_db> but I keep getting this error message:
>>>>>>>> create database "<new_RAC_db>"
>>>>>>>> *
>>>>>>>> ERROR at line 1:
>>>>>>>> ORA-01092: ORACLE instance terminated. Disconnection forced
>>>>>>>> The servers (hosted) came online yesterday with <test_RAC_db>
>>>>>>>> running (no db build script provided!)however, because of the
>>>>>>>> db_block_size (too small) this
>>>>>>>> is of no use to our business needs
>>>>>>> What is the operating system?
>>>>>>> What block size did it come with?
>>>>>>> What block size do you need and why?

>>>>>>>> create database "<new_RAC_db>"
>>>>>>>> maxinstances 10
>>>>>>>> maxlogfiles 20
>>>>>>>> maxdatafiles 100
>>>>>>>> maxlogmembers 5
>>>>>>>> character set "UTF8"
>>>>>>>> datafile '/opt/app/oracle/oradata/<new_RAC_db>/system/
>>>>>>>> system01.dbf' size 250M
>>>>>>>> <balance snipped>
>>>>>>> I have never seen anyone successfully build a RAC cluster using
>>>>>>> this method. I am sure someone has but why put yourself through
>>>>>>> so much pain with no gain.
>>>>>>> Use DBCA.
>>>>>>> --
>>>>>>> Daniel A. Morgan
>>>>>>> Oracle Ace Director & Instructor
>>>>>>> University of Washington
>>>>>>> (replace x with u to respond)
>>>>>>> Puget Sound Oracle Users Hide quoted text -
>>>>>>> - Show quoted text -
>>>>>> Hi Daniel,
>>>>>> Thanks for your quick reply.
>>>>>> The OS is Red Hat Enterprise Linux AS 4
>>>>>> The block size for <new_RAC_db> is 8192
>>>>>> The current requirement is 16384 which when created will be
>>>>>> populate by an export dmp.
>>>>>> All our tablespaces for the database being migrated across has a
>>>>>> 16k blocksize.
>>>>>> Forgive my ignorance but I have to admit I am not clear on how to
>>>>>> call up DBCA as the servers are being hosted by an outside
>>>>>> company. Thanks again for you reply.
>>>>> 8K is the correct block size for Linux. Creating a 16K blocksize
>>>>> will likely gain you precisely nothing but extra work. Look up how
>>>>> to choose a block size on Howard Roger's site:
>>>> Can you provide a more precise pointer to Howard's site where he
>>>> states "8K is the correct block size for Linux"?
>>>> Please read his article which
>>>> seems to be far more balanced than such a simplistic "silver
>>>> bullet" advice.
>>> Let me point you to the first sentence of what Howard wrote:
>>> "If you are buffering your I/O (that is, your file system has its
>>> own buffering mechanism, as most do by default), then your Oracle
>>> block size should match your file system's buffer size exactly"
>>> With every system I have ever seen where the o/s is Linux or Windows
>>> that will be 8K.
>> Do you always draw conclusions after reading the opening sentence of
>> an article?
>> And in this case you also seem to neglect the "if ..then .." nature
>> of that sentence.
>> Near the end of the article you'll find a paragraph "And if I were a
>> 10g data warehouse with direct I/O and automatic tuning" where
>> Howard explicitly favors 16K over 8K.
>> May I also suggest reading the last paragraph "So there are no
>> really simple answers?"
> Do you always make assumptions that someone is using something even
> when they don't say they are?

Excuse me?! Who is making assumptions here? There's 2 possibilities: either the OP uses IO-buffering at the OS-level or he doesn't.
You are assuming he is, given your pointer to Howard's opening sentence and your conclusion from that.
I didn't assume either way, just pointing to other possibilities (provided in Howard's article).

> Most installations are vanilla.

Perhaps, but 'most' is not 'all'.
Why do you make the assumption OP's installation is vanilla, even when he didn't say it is?
As a matter of fact, he did make some remarks to the contrary: 1. It is a RAC installation (hardly 'vanilla', I would say) 2. The existing installation on Solaris uses 16K blocksize because of business requirements

> Have you ever actually benchmarked a system with 8K vs 16K blocks?

Can't say I have, do you?

> Do it and report your results.

So you can misquote my results like you did with Howard's article?

> The OP is in the 99th percentile for wasting time.

Making assumptions again, based on what?

Received on Mon Dec 17 2007 - 17:03:13 CST

Original text of this message