|
The SAME strategy provides a clue to the fact that how a disk array is configured makes all the difference. You refer to "[t]he complexity and abstraction involved in partitioning the physical disks into the chunks presented to the servers" -- that's where you have to optimize for Oracle. In general, RAID 1+0 (mirror, then stripe) provides the best fault tolerance and performance. A database of many terabytes should employ at least several LUNs if only for maintainability -- you don't want a single huge logical volume for a database of this size. Consider providing individual logical volumes for logical divisions of this database.
I heartily recommend the paper Sane SAN by James Morle, for a discussion of SAN layout for Oracle databases.
|
