I've just taken over the architecture of an existing database that's
designed to work over six disks on Solaris systems. To improve
efficiency, the original group striped the six disks into three and
placed all tablespaces on different disks, leaving the OS to place data
on different spindles. Partitions are used, but ony to keep track of
data by a particular value so that it can be dropped easily when it is
outdated. All partitions for a database table are currently stored in
the same tablespace and on the same spindle pairs.
I don't believe that this is the optimal configuration for performance.
I would like to partition each table across multiple spindles to
improve performance and ensure that the loss of one disk does not cause
the loss of all data in a table. Can anyone tell me if partitioning
tables across mulstiple spindles will improve performance? If so, are
there available case studies that can provide me with benchmarks ofr the
percentage gain in performance?
Thanks in Advance,