Configuring RAID for best Oracle performance

How should I configure a new RAID subsystem to balance the opposing requirements of good Oracle performance and good data availablity? I understand that there are trade-offs but I'd like to learn from others' experiences and even theories so I don't have to reinvent the wheel. It would be nice to have the system keep running without physical intervention should a disk drive fail but the users (of course!) demand the best performance.

The software is Oracle-based general ledger, accounts receivable, accounts payable, human resources type of work. This is a client-serve architecture. SQLnet and PC/TCP+ are being used on Windows-based 486 PCs over token ring. The PCs also run Netware to two Netware file servers.

The server is an IBM RISC System/6000 model 590 with an IBM 7135 RAID subsystem. The 7135 has five 2 GB drives currently and can be configured with RAID 0, 1, 3 and 5. RAID configurations can be mixed (e.g., RAID 1 mirroring on two disks, RAID 3 on the other three, I think!)

Some brainstorming we've already done: 1) mirror the transaction logs using two disks and put the indexes and data across the other three. Since all the logs together will be much less than 2 GB, this seems like a waste of disk. Mirroring the data using RAID 1 also may be overkill. 2) we could use RAID 5 across all five disks and configure three logical volumes: one for indexes (near the center/fastest part of the disks?), one for transaction logs (middle part of disk), and one for data (on the rest of the disks). But then there is no control over which disk drive has the transaction logs and indexes for (manually) optimizing performance.

Any suggestions regarding RAID configuration and other potential bottlenecks would be greatly appreciated.

P.S. And a fundamental question: should we use RAID at all or just use the disks as if they were five regular drives? AIX itself can mirror disks without the use of RAID. Received on Thu Jan 05 1995 - 16:48:47 CET