RE: exp performance question ( direct=y)

From: Christopher Spence <cspence_at_FuelSpot.com>
Date: Fri, 22 Jun 2001 12:35:26 -0700
Message-ID: <F001.003344A3.20010622124458@fatcity.com>

You can expect to see as much as 75% performance gain on export. There is just about no performance gain on the import. The reason for this is it avoids 3 copies and 2 character conversions when exporting the data. This is due to the conversion from column major to row major when bringing in the buffers into the buffer cache, then converting back to column major to handle the select. I would recomend looking at the recordlength parameter as it has a good (1-4%) effect on performance initially, but after 32k it generally has very little performance gain. During conventional exports, recordlength can prove as much as 6% performance gain. Ussually setting it to 32k is the best you will see.

One thing to know about exports is conventional exports are cpu bound where as direct exports are Io bound. But either will only use a fractional portion of the cpu/io. Also note, if your using consistent exports it will incur locking as well as redo/rollback generation. So time for exports cannot be 100% accurately consistent.

Another thing is direct exports use as much as 65% less memory than conventional and about 35% less cpu. Also direct exports tend to do a three fold increase in io throughput over conventional. You can ussually export a max export sustained speed of around 5-6% of max i/o.

2) Target, source, and client used to export must be of the same character set and you cannot export lobs.

This is perhaps more than you asked for, but hopefully it helps.

"Walking on water and developing software from a specification are easy if both are frozen."

Christopher R. Spence
Oracle DBA
Fuelspot

-----Original Message-----

Sent: Friday, June 22, 2001 12:56 PM
To: Multiple recipients of list ORACLE-L

Oracle : 8.0.5
Platform : Sun

Currently we have cron job every night (starting from 11pm) to do export. I changed the setting "direct" to "y" two days ago while leaving all other parameters unchanged, hoping to gain some performance. I am a bit surprused to find that it did not. It actually took longer to create dump file with less data to export. The whole exp process takes about 2 hours to finish. Yes, there could be lots of other unix processes running during that time. But I would still expect to see some improvement because we are doing this way for quite a while. So my questions are:

From your "real" export experience, how much performance boost did you see when you set "direct=y"?
If "direct=y" improves the performance, why would anyone want to use "direct=n"?

Thanks.

Guang

here is my orcle dump file's time stamp: (dmp.1 and dmp.2 are from direct=y, dmp.3, dmp.4 and dmp.5 are from direct=n).

-rw-rw-r-- 1 mt prog 1042197132 Jun 18 01:05 oracle.dmp.5.gz
-rw-rw-r-- 1 mt prog 1042375633 Jun 19 01:04 oracle.dmp.4.gz
-rw-rw-r-- 1 mt prog 1042556662 Jun 20 00:25 oracle.dmp.3.gz
-rw-rw-r-- 1 mt prog 1034773279 Jun 21 01:17 oracle.dmp.2.gz
-rw-rw-r-- 1 mt prog 1035237986 Jun 22 01:22 oracle.dmp.1.gz

--here is the parameter file:

BUFFER = 64000
COMPRESS = Y
CONSISTENT = N
CONSTRAINTS = Y
DIRECT = Y
FILE = /oracle/exports/oracle.dmp.pipe
#FULL = Y
GRANTS = Y
INDEXES = Y
LOG = /oracle/exports/export.log
ROWS = Y
USERID = xxx/yyy
OWNER = (aaa,bbb)

Get your FREE download of MSN Explorer at http://explorer.msn.com

--

Please see the official ORACLE-L FAQ: http://www.orafaq.com
--

Author: Guang Mei
INET: zlmei_at_hotmail.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists


--------------------------------------------------------------------

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).
--

Please see the official ORACLE-L FAQ: http://www.orafaq.com
--

Author: Christopher Spence
INET: cspence_at_FuelSpot.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists


--------------------------------------------------------------------

To REMOVE yourself from this mailing list, send an E-Mail message