Veritas Storage Foundation™ for Oracle 5.0.1 Administrator's Guide

■ If you are using ODM on Oracle9, you do not need to change any init.ora
parameters.
Your Oracle Installation Guide provides detailed instructions on implementing
asynchronous I/O on your system.
Tuning buffer cache
The UNIX buffer cache plays an important role in performance when using HFS
or JFS in buffered I/O mode.
When using Quick I/O, however, the database cache must be tuned as if raw devices
are being used. You can allocate more memory to the database buffer cache because
Quick I/O bypasses the file system cache to improve database performance.
Memory pages normally allocated to the file system cache can be allocated to the
database buffer cache (SGA). With Oracle9i, you can adjust the SGA size without
shutting down the database.
Setting Oracle block reads during sequential scans
The DB_FILE_MULTIBLOCK_READ_COUNT parameter specifies the maximum number
of blocks Oracle reads in one I/O operation during a sequential scan. When the
file system is created on a striped volume, set this parameter to a value that is a
multiple of the full stripe size divided by DB_BLOCK_SIZE. Using a full stripe size
allows the read operations to take advantage of the full bandwidth of the striped
disks during sequential table scan.
Set the DB_FILE_MULTIBLOCK_READ_COUNT to a value that is a multiple of
(read_pref_io*read_nstream)/DB_BLOCK_SIZE, but the value should not exceed
the value of max_direct_iosz/DB_BLOCK_SIZE.
Use the vxtunefs command to display the value of read_pref_io, read_nstream,
and max_direct_iosz, for example:
# vxtunefs /db01
The vxtunefs command displays output similar to the following:
Filesystem i/o parameters for /db01
read_pref_io = 65536
read_nstream = 4
read_unit_io = 65536
write_pref_io = 65536
write_nstream = 4
write_unit_io = 65536
313Tuning for performance
About tuning Oracle databases