Re: Oracle Linux Cluster File System

In article <aheeq1$eta$00$1_at_news.t-online.com>, "Norbert says...
>
>Sorry Pete. Miss-spelled one-off. I tried to say I had to create 8
>tablespaces only once, 3 tablespaces per node (2x redo, 1x undo). That's why
>I thought I'll need 14. And TEMP is per node too?! Would change the sum to 7
>+ 2*4 = 15...
>
>Norbert

Ah hah! Now I understand what you meant. You confused me by throwing in the term tablespace when you mean redo logs. Redo logs and control files are not tablespaces. Tablespaces are logical structures that are represented on disk by datafiles. So going back to your original file listing, the datafiles that are shared between nodes are for these tablespaces:

system (10GB)
indx (50GB)
user (50GB)
tools (10GB)

You can also have these set up per node if you want to:

temp_nodeX (10GB)
undo_nodeX (10GB)

So you may end up with 8 tablespaces, and however many redo logs and control files you want, based on your requirements from both an availability perspective and a performance perspective for the redo logs.

Pete
>
>"Pete Sharman" <peter.sharman_at_oracle.com> schrieb im Newsbeitrag
>news:ahddqk02gaj_at_drn.newsguy.com...
>> In article <ahbcfv$lhj$03$1_at_news.t-online.com>, "Norbert says...
>> >
>> >Pete,
>> >
>> >> as I suspect there's a fundamental lack of understanding here
>> >
>> >Oh, I'm sorry about that! Let me explain.
>> >
>> >Fundamental Breakdown: Let's not discuss about the suggested sizes.
>> >
>> >Second point: mknod does not create a tablespace, it creates a special
>file.
>> >raw does not create a tablespace, it binds a raw device to block devices.
>> >So, I did not want to create physical files for every node.
>> >
>> >Before going any further: Let's say I issue the fdisk command on one node
>> >only (which I intended to do anyway).
>> >
>> >I really would like to get some comments on my scriptlets. Or at least
>not
>> >only a RTFM
>> >suggestion without a URL.
>> >
>> >Thanks in advance,
>> >
>> >Norbert
>>
>> Norbert
>>
>> Somebody else already responded on that posting for it, and since I'm not
>a deep
>> Linux expert I didn't have anything to add of value to that. The
>fundamental
>> comment I was making (apart from the sizes which are still crazy) was to
>your
>> statement in the original posting:
>>
>> "that's 8 tablespaces on-off, 3 per node, 2 nodes make 6+8 = 14
>tablespaces,
>> size = 160 GB + 2 * 3 * 10 GB = 220 GB)"
>>
>> I can't understand what you're saying here. What do you mean by 8
>tablespaces
>> on-off? Tablespaces aren't per node except the UNDO and TEMP tablespaces
>(and
>> even there they are per instance not per node), so the 3 per node doesn't
>make
>> any sense, and then the math totally throws me. Can you explain what you
>mean
>> by this part?
>>
>> Pete
>> >
>> >
>> >"Pete Sharman" <peter.sharman_at_oracle.com> schrieb im Newsbeitrag
>> >news:ah9d8u01kb1_at_drn.newsguy.com...
>> >> In article <ah8ok2$aro$03$1_at_news.t-online.com>, "Norbert says...
>> >> >
>> >> >Since you mentioned "sales":
>> >> >
>> >> >I would be pleased to get some helping hand with regard to raw devices
>> >then:
>> >> >
>> >> >(see RE: RAC on Linux Cluster, Message-ID:
>> >> ><ah6tgd$spt$04$1_at_news.t-online.com>)
>> >> >
>> >> >Please comment the suggested prodecure (I start with two nodes)
>> >> >
>> >> >The following tablespaces are needed:
>> >> >
>> >> >system (10GB)
>> >> >indx (50GB)
>> >> >temp (10GB)
>> >> >user (50GB)
>> >> >tools (10GB)
>> >> >undo_nodeX (10GB)
>> >> >control1 (10GB)
>> >> >control2 (10GB)
>> >> >control3 (10GB)
>> >> >redo1_nodeX (10GB)
>> >> >redo2_nodeX (10GB)
>> >> >
>> >> >(that's 8 tablespaces on-off, 3 per node, 2 nodes make 6+8 = 14
>> >tablespaces,
>> >> >size = 160 GB + 2 * 3 * 10 GB = 220 GB)
>> >>
>> >> OK, fundamental breakdown in understanding right here. For one thing,
>> >your
>> >> files are ridiculously large. Why are you setting up everything in
>chunks
>> >of 10
>> >> Gb? 10 gig is far, far too large for any control files or redo log
>files,
>> >and
>> >> unless you have a ginormous database with literally trillions of lines
>of
>> >PL/SQL
>> >> code you won't need a 10 gig SYSTEM tablespace either. Likewise the
>other
>> >files
>> >> are for tablespaces that are used generally in demo systems, and I've
>> >never seen
>> >> a demo database 190 Gb in size! :)
>> >>
>> >> Second point, in RAC the datafiles are shared i.e. accessible from each
>> >node,
>> >> but there's only one copy of each file e.g. you don't have a SYSTEM
>> >tablespace
>> >> on each node, unless you have multiple databases.
>> >>
>> >> Before you go any further, I strongly suggest you read both the Oracle
>and
>> >RAC
>> >> Concepts manuals, as I suspect there's a fundamental lack of
>understanding
>> >here.
>> >>
>> >> Pete
>> >> >
>> >> >I would do it like:
>> >> >
>> >> >on node1:
>> >> >
>> >> >fdisk /dev/sda
>> >> >
>> >> >create first 4 partitions primary
>> >> >then one big extended
>> >> >then therein 14-4 logical partitions
>> >> >
>> >> >(this creates devices /dev/sda1 ... /dev/sda14
>> >> >
>> >> >on every node:
>> >> >
>> >> >create special files for every tablespace
>> >> >
>> >> >mknod /dev/system.dbf c 162 1
>> >> >...
>> >> >
>> >> >bind the block device to the raw device
>> >> >
>> >> >e.g.:
>> >> >
>> >> >raw /dev/system.dbf /dev/sda/sda1
>> >> >
>> >> >instead, write the config file (suse, package rawio)
>> >> >
>> >> >system.dbf:sda1
>> >> >indx.dbf:sda2
>> >> >...
>> >> >
>> >> >edit /etc/rc.config
>> >> >
>> >> >ACTIVATE_RAW_DEV=yes
>> >> >
>> >> >
>> >> >chown oracle /dev/system.dbf
>> >> >
>> >> >etc.
>> >> >
>> >> >
>> >> >Thanks in advance,
>> >> >
>> >> >Norbert
>> >> >
>> >> >
>> >> >"Pete Sharman" <peter.sharman_at_oracle.com> schrieb im Newsbeitrag
>> >> >news:ah76nq02q45_at_drn.newsguy.com...
>> >> >> In article <3D371161.13DD_at_yahoo.com>, Connor says...
>> >> >> >
>> >> >> >Pete Sharman wrote:
>> >> >> >>
>> >> >> >> In article <ah6tk1$elf$01$1_at_news.t-online.com>, "Norbert says...
>> >> >> >> >
>> >> >> >> >At OracleWorld Copenhagen Oracle said they will release a CFS
>for
>> >> >linux this
>> >> >> >> >month.
>> >> >> >> >
>> >> >> >> >Does anybody know something about actual release date?
>> >> >> >> >
>> >> >> >> >Thanks,
>> >> >> >> >
>> >> >> >> >Norbert
>> >> >> >> >
>> >> >> >> >
>> >> >> >> Yes. Can I tell you? No.
>> >> >> >>
>> >> >> >> Sorry, couldn't resist! :)
>> >> >> >>
>> >> >> >> HTH. Additions and corrections welcome.
>> >> >> >>
>> >> >> >> Pete
>> >> >> >>
>> >> >> >> SELECT standard_disclaimer, witty_remark FROM
>company_requirements;
>> >> >> >
>> >> >> >(and since you're an Oracle employee...)
>> >> >> >
>> >> >> >even if you did tell us - would be it correct tomorrow ? No!
>> >> >> >
>> >> >> >:-)
>> >> >>
>> >> >> Oohhh, cutting, cutting! Do you see the word "Sales" in my title?
>:)
>> >> >>
>> >> >> Pete
>> >> >> >
>> >> >> >--
>> >> >> >==============================
>> >> >> >Connor McDonald
>> >> >> >
>> >> >> >http://www.oracledba.co.uk
>> >> >> >
>> >> >> >"Some days you're the pigeon, some days you're the statue..."
>> >> >>
>> >> >> HTH. Additions and corrections welcome.
>> >> >>
>> >> >> Pete
>> >> >>
>> >> >> SELECT standard_disclaimer, witty_remark FROM company_requirements;
>> >> >>
>> >> >
>> >> >
>> >>
>> >> HTH. Additions and corrections welcome.
>> >>
>> >> Pete
>> >>
>> >> SELECT standard_disclaimer, witty_remark FROM company_requirements;
>> >>
>> >
>> >
>>
>> HTH. Additions and corrections welcome.
>>
>> Pete
>>
>> SELECT standard_disclaimer, witty_remark FROM company_requirements;
>>
>
>

HTH. Additions and corrections welcome.

Pete

SELECT standard_disclaimer, witty_remark FROM company_requirements; Received on Sun Jul 21 2002 - 22:46:26 CDT