Re: I/O Contention Oracle binaries, DB File Placement on 4 mount points

David,

I don't have the link handy, but I can recall that for (throughput - overhead) : a 4 drive RAID 5 array ~ a 2 drive raid 1 array (lots of averaging going on there).

You don't want the system datafile on a JBOD (just a bunch of disks - non-RAID) drive under any of the combinations. HA won't get you around having to recover the database with a restored copy of the system datafile (probably lots of archived redo logs to apply to get back up to date).

the online redo logs need not be on a RAID 1 set, could duplex logs on separate disks (JBOD).
Oracle uses all log members in a group to feed the archiver in a round-robin fashion - so multiple members on non-raid disks is better than 1 member on a RAID 1 volume. Helps prevent corruption. Others here would say use 3 log members per redo log group (Howard?).

again, can you survey this database running elsewhere to find out the I/O profile?
volume of redo generated per working day (arch logs volume) I/O load vs. time of day (plan your backups) number of reads/writes (98% read on data, index?) bytes read/written by file, tablespace (look for large values for RBS, TEMP) If you have lots of trans, you'll want RBS on a RAID 1 volume. Where are archived redo logs stored in these configs again?

One more thing to keep in mind - its nice to have enough space to be able to do a cold backup prior to performing recovery - if its going to be hairy - although others would say - provided you backup your control files (cold) prior to starting recovery - you're still okay.

Paul

David Heitholt wrote:
>
> Thanks for the response, Paul.
>
> I left out some info - the production db server is setup for HA. The
> db server box (hosting the Oracle binaries) is completely mirrored by
> a backup box. In case of failure in the primary box, HA fails over to
> the backup box. One of the mount points for Oracle is on this server
> with HA (the mount point with the Oracle binaries). All other mount
> points are (one way or another) on the 9 disk data array. My SA would
> like to keep one of the 9 disks there as a spare disk, thus leaving 8
> disks. The 9 disk array has 2 controllers. I want to insist on getting
> an additional drive for archived redo logfiles and hot backup
> datafiles – meaning that for my drives for backup are in
> question.
>
> I have a concern for keeping DATA and INDEX on 2 separate logical
> volumes so that I know I can get parallel reads of those volumes. So
> if I take the 2 (3 disk) RAID5 mount points into one RAID5 mount
> point, I end up putting DATA and INDEX on that one mount point.
>
> I also think I should keep RBS and TEMP tablespaces on a drive
> separate from DATA and INDEX.
>
> Mount points as per my original plan:
>
> /u01 1 disk (mirrored via HA) - Oracle binaries, system, temp,
> rbs users, tools tablespaces, controlfile copy
> /u02 2 disks in RAID1 volume online redo logs
> /u03 3 disks in RAID5 volume - DATA, controlfile copy
> /u04 3 disks in RAID5 volume - INDEX, controlfile copy
>
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Alternative Configuration 1:
>
> /u01 1 disk (mirrored via HA) Oracle binaries, controlfile copy
> /u02 (disk array) 2 disks in RAID1 volume - online redo logs
> /u03 (disk array) 2 disks in RAID1 volume - DATA, controlfile copy
> /u04 (disk array) 2 disks in RAID1 volume - INDEX, controlfile copy
> /u05 (disk array) 2 disks in RAID1 volume - system, temp, rbs users,
> tools tablespaces and more DATA datafiles as the space is
> needed for data.
>
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Alternative Configuration 2:
>
> /u01 1 disk (mirrored via HA) - Oracle binaries, system, temp,
> rbs users, tools tablespaces, controlfile copy
> /u02 (disk array) 2 disks in RAID1 volume - online redo logs
> /u03 (disk array) 6 disks in RAID5 volume - DATA, INDEX,
> controlfile copy
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> 3 drive RAID 5 volumes are useless.; - Why is that ?
> My reason for going with 2 RAID5 volumes (each of 3 drive) is getting
> more space out of the drives and the separation of volumes for DATA
> and INDEX.
>
> What is JBOD single drive?
>
> Thanks in advance.
>
Received on Sat Sep 22 2001 - 22:51:17 CDT