oracle 8i on aix & filesystems

[smacattack]

Is there a standard setup for creating a oracle database on aix filesystems.
Should there be a filesystem setup for say datafiles , rollback segment, indexes , archive logs etc or should it be 1 filesystem with everything thrown in?
Can anybody indicate any preference on new aix hardware and oracle 8i.
We are use to old hardware with 6 oracle just a bit different!
cheers

 

[gheist]

AIX offers raw disk partitions in place of every file oracle uses to store data, and you can put it on the whichever disks you want

You should take ask people who sell you things, as Oracle server will start and continue to run on even underpowered machines tricked to run top modern operating system on older hardware

 

[anshudad]

There is no standard setup defined. But for better performance your datafiles and indexfiles should be in separate filesystems(actually separate disks). So when you create the logical volume and filesystems you have to choose different disks for those filesystems.

 

[mrn]

Hi, There is a sort of standard, as used by oracle engineers

/u01/oracle - contains oracle software
/u02/oradata - for your datafiles
/u03/redo - For redo logs
/u04/archive - archive logs

all on individual disks (mirrored if poss)

if installing multiple oracle instances, carry on with

/u11/oracle
/u12/oradata
/u13/redo
.....

when installing oracle will create sub-directories under

/u01/oracle/SIDNAME/......

Regards Mike --
| Mike Nixon
| Unix Admin
|

http://www.instantflashshop.com

----------------------------

 

[Yegolev]

We have two different methods. The easy one is one-datafile-per-fs and one-fs-per-disk. The hard one is one-datafile-per-fs and eight-fs-per-eight-disks. The second one is to help spread the love amongst several DASD for I/O reasons and is not exactly striping, but we sequentially allocate the PPs one at a time to eight disks.

You can have loads of fun with this stuff. =)

 

[gheist]

some Oracle can use raw logical volumes (partitions,via aio) , blah blah blah ...

yes - indices can be separated from data, but this takes indepth analysis of database design and applies to every platform

the more procesors the better (db performance = (perf on 1 cpu)*(numberofcpus-1) so minimum is 3 CPUs (or 2 RS64 split via HMT into 4)
, etc etc


I know of two critical bottlenecks
- disks cannot write at reasonable speed so db or whole system crashes - assume buying disks with reasonable speed rating

- previos fills all the ram and excessive paging brings sistem nearly down
assume buying sufficient memory for your database server (ie [memory used now]x1.5[future growth]x2[Expectations from new software version])

bottlenecks will be somewhere else anyway - as same database is already running on older hardware.

 

[bi]

It is not only the number of disks you have to consider in avoiding bottlenecks, but also IO cards. If all your disks are on the same card, performance will suffer.

Yegolev, have you considered using the maximum range of physical volumes option when creating the logical volume
(-e x)? You would name all the physical volumes you want to use and AIX sequentially spreads the LV across all the disks. This would be a little faster than allocating one partition at a time.

 

[Yegolev]

The reason we do it the way we do is to put the first PP on a different disk each time. 8 fs, 8 DASD, each DASD has PP #1 of a different fs.

fsA = pp1-disk1, pp2-disk2, etc
fsB = pp1-disk2, pp2-disk3, etc
fsC = pp1-disk3, pp2-disk4, etc
etc

I did not develop the method but I understand it involves undocumented commands for allocating PPs. Doing it your way will work fine but I was told it can be somewhat random, and the guy who came up with this is the type who would rather be sure than fast. =) I think it should do it sequentially, like you said, and maybe I am just thinking of a different thing. I can look up the script if you want... I will probably look through it myself anyway now that I am curious about it. As usual I could be wrong.

As for adapter cards, we use dual fibre HBAs to give two paths to the EMC disks. It has been a while since I talked about this with the Storage team but I believe it does load-balancing as well as provide redundancy. I can look that up, too, when I get back in the office.