At this time, we are not aware of any specific effect caused by running with different block sizes. 8 K blocks, in particular, should be fine. Due to the way in which the database is primarily used, larger block sizes are not expected to perform better. For information on how the database is primarily used, see How the Management Server Uses the Database .
Oracle® can utilize striping to increase the throughput of the database. A responsible Oracle DBA will analyze the I/O numbers of our sizing spreadsheet and then set up striping if they feel that faster disk access is warranted. In testing, we have not seen any need for striping for the supported load numbers. Use striping at your own discretion.
Foglight® Management Server basically manages two sets of tables:
The configuration of multiple tablespaces enables the Management Server to spread the observation tables over multiple hard drives (if configured that way) and, generally, to achieve higher throughput for incoming data, retrieval, and rollup operations. At present, we are not aware of a bottleneck in Oracle® performance with the supported load numbers and known hardware. For higher supported load numbers, and, depending on the hardware, this setup should be considered by the DBA.
For more information, consult Oracle’s documentation on tablespaces.
With Oracle9i and beyond, the system should be able to accomplish what it needs to without you having to add rollback segments, and it should be able to go as far back in time as the UNDO_RETENTION parameter specifies. When UNDO_RETENTION is equal to 900 (the default value), the undo retention time is 15 minutes. If the database is transaction-intensive (as is the case with Foglight®), Oracle9i creates a lot of rollback segments for each transaction on the undo table space. When the transaction is finished, it keeps the rollback segments to satisfy the UNDO_RETENTION value. To be able to serve new transactions, it creates new rollback segments. If the autoextend parameter is on, the UNDO_TABLESPACE keeps growing. In comparison with Oracle8i, if UNDO_RETENTION=20, the Oracle9i UNDO_TABLESPACE size has to be 20 times more than all rollback segments in Oracle8i, given the same transaction rate. It is possible to have a fixed UNDO_TABLESPACE size, like in Oracle8i. In that case, it has to initially be large enough to be able to process all transactions, and, at the same time, keep all used rollback segments to satisfy the UNDO_RETENTION. So, in our case, we need to reduce UNDO_RETENTION and estimate the transaction rate. Oracle recommends that you use the following formula:
UndoSpace = UR * UPS + overhead
• |
UndoSpace is the number of undo blocks |
• |
UR is the UNDO_RETENTION value in seconds |
• |
UPS is the number of undo blocks per second, or the transaction rate |
• |
overhead is the small overhead (disk space) for the metadata (transaction tables, bitmaps, and so on) |
Example instructions (full set):
(for example, NEWUNDOTABLESPACENAME] = UNDOTBS2, [TABLESPACEDATAFILE] = /data02/oracle10g/oradata/FGL/undo0201.dbf, [OLDUNDOTABLESPACENAME] = UNDOTBS1 )
© ALL RIGHTS RESERVED. Feedback Terms of Use Privacy Cookie Preference Center