| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Usenet -> c.d.o.server -> disk i/o balancing
We're debating different ways of optimizing disk i/o. The ground rules: we WILL be using some sort of redundancy for everything. Everything but redo logs will probably be on RAID 1 or 0+1, or RAID 0 with LVM mirroring; redo logs will probably be duplexed by the RDBMS onto multiple plain old drives. Of course, tables & indexes will be separate.
One side would like to split our tables and indexes into two tablespaces (and RAID sets) each. Typically, parent/child tables (and other tables known to be used together often) will be split, thus spreading out the i/o over four RAID sets (assuming indexes are used for both tables in a two-table join).
The other side doubts that we can choose that wisely. They say that most queries involve some number of tables other than two; that what's a parent in one query is a child in another; that the applications will change over time to make the original choices bad ones; that this is a high-maintenance setup. This side advocates fewer but larger RAID/LVM sets, over more disks & more SCSI channels. Distribute the i/o widely at and below the file system level, and trickle down those advantages to Oracle. Break up tablespaces at the datafile level if need be.
What do you think?
Thanks
Joseph
O- Received on Wed Feb 05 1997 - 00:00:00 CST
 |
 |