Operating System - HP-UX
1748227 Members
4395 Online
108759 Solutions
New Discussion юеВ

More hd layout for oracle8 db questions

 
SOLVED
Go to solution
Bert Woods
Frequent Advisor

More hd layout for oracle8 db questions

Hi,

I just saw the previously posted question. But, I though mine
was a little more specific. I'm going to read up on this where
suggestions are posted. Thanks for looking at this though...

I have four controller connected to three drives each
and I am trying to lay out the drives in the most efferent
manner and to keep the data protected. Oh, yeah this is a
Oracle 8 db


1 4 = db engine
2 5 |
3 6 |
----------- = data, index, redo, system
a d |
b e |
c f = c - archive, f - rollback


My idea is this:
the first two drives (1,4) are mirrored and contain the db engine
the (c) drive is used for archive (unmirrored and unprotected)
the (f) drive is used for rollback segments (all three) (unmirrored and unprotected)
(2,3,a,b) are striped
(4,5,d,e) are striped
(2,3,a,b) and (4,5,d,e) are mirrored

user data, indexes, redo, system tables are spared across the mirrored
(2,3,a,b) and (4,5,d,e)

Have I confused you and if not: what do you think?

I appreciate all replies

"it's not a problem until it a problem"
5 REPLIES 5
Dave Chamberlin
Trusted Contributor

Re: More hd layout for oracle8 db questions

If what you call archive is the Oracle archived log files - I would recommend that these be mirrored. I would also mirror the rollback segments. Tables and Indexes should not be on the same disk (for the same schema).

Re: More hd layout for oracle8 db questions

Without commenting on the rest of your plan, I would point out one issue which I think you should know about...

Definately mirror your archived redo logs - these are absolutley *the* most important part of your data - mirroring your data files etc. might protect you from a physical corruption (when a disk fails) but *only* your archived redo is going to protect your data from a *logical* corruption.

If you talk to Oracle about 'guaranteed writes' they will say something along the lines of "we guarantee that on commit data is written to redo, we don't guarantee it gets written to your datafile"

Protect your archived redo at least as well as you protect your datafiles.

HTH

Duncan

I am an HPE Employee
Accept or Kudo
Tim D Fulford
Honored Contributor
Solution

Re: More hd layout for oracle8 db questions

Read the SAME paper.

http://forums.itrc.hp.com/cm/components/FileAttachment/0,,0x4a8a8cc5e03fd6118fff0090279cd0f9,00.pdf

you only have a few disks, I assume they are the same size. the best (generic) performance would come from Stripe And Mirror Everything. You have 10 disks I believe, so split them into two groups of 5 on different controllers. Then create a /etc/lvmpvg with the two sets of disks (man lvmpvg) then
# lvcreate -L -n -D y -s g -m 1 vgdb

If you try to use functional volume groups (i.e. disks for indexes, data, redologs, rolback, archive....) you will find than you may run into performance or capacity problems quicker than you planned...

Tim
-
Bert Woods
Frequent Advisor

Re: More hd layout for oracle8 db questions


Wonderful, this was what I was looking for!

Ok, they are all 9 gig hvd drives (total of 12, not including the mirrored two for the OS) on
two Jamaica boxes and four controllers (3 drives per)

The part with (4,5,d,e) should read (5,6,d,e) (4) is being used by the db engine mirror, oops.

Ok, so it's a bad idea to have the rollback and archive log (yes that is what I meant) w/o protection? The are on
separate drives because if I recall correctly the rollbacks are copied to the archive logs when they fill.
I thought as long as the redo logs & data or the ARCHIVE logs was mirrored/raided I was ok. Didn't need both.

I'm stuck with the number of drives I have so I'm trying to get the most out of the existing system.

I currently have the drives split into function groups (as Tim offers) I noticed the data and index mirrors are
hit the most and the system, redo, rollback, user data are seldom accessed. So, what Tim offered is (I guess) what
I was looking for. Just didn't know if it was a "BAD" idea.

I am reading the SAME docs, now.

I still am looking any input...


Thanks!!!!
"it's not a problem until it a problem"
Volker Borowski
Honored Contributor

Re: More hd layout for oracle8 db questions

Hi,

well, not completely.

ROLLBACK refers to rollback segments and is a part of the database and kept in tablespaces, so it should be among your SYSTEM / DATA / INDEXES.

"redo" refers to online redo logs. These should be on mirrored disks or can be mirrored with oracle tools AND should not be located among archive.

To be safer, I would mirror c/f for archives as well (otherwise if the archive disk goes down, your database will stop a little later!).

Put redo onto your 1/4 disk (among the executables) and mirror them with oracle to your data/index area.
ALTER DATABASE ADD LOGFILE MEMEBER '...' TO GROUP n;

This will be safer, but may cost performance, because redo,rollback and data are on the same logical disks.

If you have swap and system elsewhere, you are fine. Otherwise, consider to mirror 2/5 to seperate OS and swap.

Hope this helps
Volker