Re: slow imports

From: Ronald <ronr_at_wxs.nl>
Date: Wed, 23 Dec 1998 20:35:59 +0100
Message-ID: <01HW.B2A704AF0004670B127E4B70_at_news.wxs.nl>


On Tue, 22 Dec 1998 22:26:27 +0100, Lisa Lewis wrote (in message <75p2ug$9lg$1_at_usenet01.srv.cis.pitt.edu>):
>
> Same old problem .... slow imports
>
> Is it wise to drop the PK index of the table (tables already exist before
> import) before importing. The documentation states that even if you
> specify INDEXES=N for the import, the unique indexes are still created.
>
> Also, should I make 1 huge rollback segment for the import or set COMMIT
> and BUFFER parameters. What would be a reasonable BUFFER setting?
>
> My imports are very slow!! Also, I noticed by looking at the alert.log
> file that I am doing a log switch very frequently. My redo logs are 10Megs.
> Is this unusually small? Should they be more on the magnitude of 100M?
>
> By the way, I am doing direct mode exports. Platform: Solaris 2.6
>
> Thanks for any insight! Lisa
>
>
Hi Lisa,

you can win by a factor of 8 by dropping the indexes and import without the indexes and constraints.

imp constraints=n indexes=n etc.

The direct mode exports only speeds up the export, not the import. I prefer to have only one BIG RBS online when possible when performing BIG imports. Normally I use buffer=64000

What is slow any way ? I import a 4G database in 2 hours, including the indexes build. When building the indexes during the import it will atleast be 8.

when possible try to split the expdat.dmp per tablespace, assuming you put each tablespace on a seperate disk this can win some time during the import (parallel) too.

merry x-mas !

-- 
Ronald  -  Unix SA                    
http://home.wxs.nl/~ronr/professional.html
        -  Oracle DBA                 http://www.tresco.nl  
        -  Middleware programmer/
                      designer        mailto:R.Rood_at_tresco.nl
 
Received on Wed Dec 23 1998 - 20:35:59 CET

Original text of this message