System level
------------------

use dedicated disks.
Move the redolog files from RAID 5 devices. RAID 5 is not efficient for writes.
Alternate redo logs on different disks to minimize the effect of the archiver on the log writer

Topic: "For Your Notepad"

Set the initialization parameter LOG_CHECKPOINT_INTERVAL to a value n (where n is an integer) to require that the checkpoint position never follow the most recent redo block by more than n blocks. In other words, at most n redo blocks can exist between the checkpoint position and the last block written to the redo log. In effect, you are limiting the amount of redo blocks that can exist between the checkpoint and the end of the log.

Oracle limits the maximum value of LOG_CHECKPOINT_INTERVAL to 90% of the smallest log to ensure that the checkpoint advances into the current log before that log fills and a log switch is attempted.

LOG_CHECKPOINT_INTERVAL is specified in redo blocks. Redo blocks are the same size as operating system blocks. Use the LOG_FILE_SIZE_REDO_BLKS column in V$INSTANCE_RECOVERY to see the number of redo blocks corresponding to 90% of the size of the smallest log file.

To calculate performance overhead, use the V$SYSSTAT view. For example, assume that you execute the following query:

SELECT NAME, VALUE
FROM V$SYSSTAT
WHERE NAME IN ('physical reads','physical writes',
'physical writes non checkpoint');

Calculate the percentage of extra I/Os generated by fast-start checkpointing using this formula:

[((PW_2 - PW_1) - (PWNC_2 - PWNC_1)) / ((PR_2 - PR_1) + (PW_2 - PW_1))] x 100% = EIO

It can take some time for database statistics to stabilize after instance startup or dynamic initialization parameter modification. After such events, wait until all blocks age out of the buffer cache at least once before taking measurements. If the percentage of extra I/Os is too high, then increase the value of FAST_START_MTTR_TARGET.

The number of extra writes caused by setting FAST_START_MTTR_TARGET to a nonzero value is application-dependent; it is not dependent on cache size.

Sorry.. No raschety verny i dla Oracle 8i ( high/8.1.5).