File Placement Across 3 RAID 1 Sets

[twood]

I currently have a 9i database on a dual Xeon machine with 4GB RAM. There are only two 10k rpm disks on a U320 SCSI controller (which provides RAID 0 or RAID 1 only). The two disks are set up in a RAID 1 which stores OS, Oracle_Home, and all data files. I have room to add 4 more disks. Now, I'm trying to decide how to use the additional disks. Iniitally, I'm thinking make two more RAID 1 sets. Keep OS and Oracle software on the existing set, move data files to the second set, and indexes to the third.

Any advice on where to put redo, undo, temporary tablspace, and other data? I'm sure there are objective metrics to help me make this decision, any advice on where to find this info?

Thanks!
-MT

[slimdave]

I would add the four new disks as a single raid 1 set, and put both indexes and tables on there.

[marist89]

Advice? I got lots of advice. My application may perform different than your application so metrics are irrelavent.

I wouldn't waste a mirrored pair by dedicating it to the OS and the Oracle Software. Ideally, I would setup three pairs (d0, d1, d2) and mirror across two controllers(c0,c1).

With only three devices, I wouldn't worry about seperating my data and indexes unless there was serious I/O contention for one of the indexes. I would spread my data AND indexes out on all three devices in LMTs with system managed extents. I would do the same with my undo. I would have two members per log group writing to d0 and d1 and use d2 as my log_archive_dest. Depending on how much TEMP is used, I might make three temp tablespaces and assign a third of my users to each.

[marist89]

Originally posted by slimdave
I would add the four new disks as a single raid 1 set, and put both indexes and tables on there.

I would love to know why you would quadruple mirror one disk.

[slimdave]

Originally posted by marist89
I would love to know why you would quadruple mirror one disk.

Ha ha. Oh yes. My bad, it must be a Friday or something. Or a Monday. Or Tuesday.

[waitecj]

If you have six disk drive bays, you could have one disk drive (not mirrored) for the OS and RAID5 the rest.

[twood]

Originally posted by marist89
Advice? I got lots of advice. My application may perform different than your application so metrics are irrelavent.

I wouldn't waste a mirrored pair by dedicating it to the OS and the Oracle Software. Ideally, I would setup three pairs (d0, d1, d2) and mirror across two controllers(c0,c1).

With only three devices, I wouldn't worry about seperating my data and indexes unless there was serious I/O contention for one of the indexes. I would spread my data AND indexes out on all three devices in LMTs with system managed extents. I would do the same with my undo. I would have two members per log group writing to d0 and d1 and use d2 as my log_archive_dest. Depending on how much TEMP is used, I might make three temp tablespaces and assign a third of my users to each.

All sounds good, except I'm wondering about the redo logs. Do I really need two members per group if they are being written to mirrored disks?

[marist89]

Originally posted by twood
All sounds good, except I'm wondering about the redo logs. Do I really need two members per group if they are being written to mirrored disks?

IMHO, Oh yeah. In case you have logical corruption of the redo log or somebody does an "rm redo1.log" you will have another multiplexed copy.