Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Mailing Lists -> Oracle-L -> RE: Commit boundary <-> Stripe Unit Size Co-relation
Vivek - I just wanted to make sure that your tests accurately reflect your
production situation. The easiest tests to construct are the simplest ones.
I've seen people draw wrong conclusions from those tests and even widely
publish those conclusions. :-(
Your original question just asked how commits related to disk writes. I just wanted to make sure you understood something about the method Oracle uses to decide when to write a block to disk. If you are just running a batch job that pounds inserts into Oracle, that is a really different situation from having many users and jobs that are doing many different things at the same time. My experience has been that those other activities very strongly affect Oracle's pattern of writing inserts and updates to disk. Mainly I have seen the insert or update job slow down a lot because it must wait for free disk blocks. Make sure you are measuring the wait statistics as you try these different tests.
This also relates to the "tuning" of your disk subsystem by setting the stripe size. If you are just doing continual writes or updates, then it makes sense that a larger stripe size may be more efficient. And if that truly represents your production environment, go for it.
There are also interactions between database block size, the operating system block size, and the disk subsystem stripe size. I have seen that discussed on this list, but I have no personal experience. ;-)
Dennis Williams
DBA, 40%OCP
Lifetouch, Inc.
dwilliams_at_lifetouch.com
-----Original Message-----
Sent: Thursday, February 13, 2003 12:39 AM
To: Multiple recipients of list ORACLE-L
Dennis , Connor , List
Further a very vague Qs.
For Batch Jobs , we get extremely DIFFERENT performances when using DIFFERENT Stripe Unit Sizes of 4K , 64 K , 128K , 512K , 1M , 2M WITH 128K performing the BEST.
Both SELECTs & UPDATEs Hang almost indefinitely with 512K , 1M , 2M Stripe Unit Sizes .
Average Size Data Inserted/Updated is 5K approx. acrross 7 Tables for one Application Transaction
Thus there Seems Some Co-relation between SELECT/UPDATE/INSERT from Datafiles & Different Underlying Stripe Unit Sizes .
Any Comments ?
Thanks
-----Original Message-----
Sent: Wednesday, February 12, 2003 8:54 PM
To: Multiple recipients of list ORACLE-L
Vivek - Just to add to Connor's statements (wow am I being rash here), Oracle's strength is that it's architecture disconnects transactions from disk writes. On one hand, block may be modified several times before being written to disk (hot block, for instance). On the other hand, Oracle may need buffer space and write a block to disk before a transaction commits. But Oracle keeps track of all this and can straighten everything out if the transaction is rolled back or the system crashes.
Dennis Williams
DBA, 40%OCP
Lifetouch, Inc.
dwilliams_at_lifetouch.com
-----Original Message-----
Sent: Wednesday, February 12, 2003 12:04 AM
To: Multiple recipients of list ORACLE-L
CASE - If Size of 1 INSERT/UPDATE Statement = 1K & Stripe Unit Size is 128 K ?
How will 1 COMMIT issued after 300 "1K" INSERT/UPDATE Statements DIFFER from 1 COMMIT issued after EACH "1K" INSERT/UPDATE Statement with respect to Writing to the datafiles on the Underlying Striped Volume ?
If 1 INSERT Statement Data is Written to the 1st Disk (say) of the Striped
Volume , will a Repeat of the SAME INSERT Statement Write to a Different
Underlying Disk of the same Striped Volume within the SAME Segment Extent ?
--
Please see the official ORACLE-L FAQ: http://www.orafaq.net
--
Author: VIVEK_SHARMA
INET: VIVEK_SHARMA_at_infosys.com
Fat City Network Services -- 858-538-5051 http://www.fatcity.com San Diego, California -- Mailing list and web hosting services ---------------------------------------------------------------------To REMOVE yourself from this mailing list, send an E-Mail message to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).
Fat City Network Services -- 858-538-5051 http://www.fatcity.com San Diego, California -- Mailing list and web hosting services ---------------------------------------------------------------------To REMOVE yourself from this mailing list, send an E-Mail message to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing). Received on Thu Feb 13 2003 - 09:04:01 CST