unicode performance penalty

I'm in the process of migrating our database to Unicode and have discovered that it works 4 to 5 times faster than non-unicode database.

Oracle docs says that Unicode db should perform same or better than a non-unicode one. Perhaps someone can verify my configuration and point possible problem areas ?

Server : Oracle 9.2.0.1.0

Client : java thin jdbc ( ojdbc14.jar )

We did not change anything in the client ( not a single line ! ); hardware and memory configuration haven't changed either.

We need to support multiple language in the same database. Most of the european languages need to be supported, but not japanese, arabic, korean or chinese.

99% of the database activity is the single insert :

insert into table1( a, b, c, d, data ) values( ?,?,?,?,? ); a,b,c,d are number(11)
data was varchar2(1024)

we execute INSERTS using jdbc batch statement ( 1000 inserts in one batch ). Autocommit is disabled.
current size of the databse is about 3 GB; expected to grow by 3GB per year. 99% of the database space is taken by table1.

Before the conversion to Unicode we could perform about 6000 inserts per second.
After the conversion we can perform only about 1800-2000 inserts per second, even if I triple RAM used by Oracle.

The database needs to be in AL32UTF8 - this is a given.

I tried using both varchar2( 1024 byte ) and nvarchar2(512) with the AL16UTF16 character set, but It didn't increase database's speed.

Any suggestions are welcome. Received on Mon Jan 08 2007 - 07:06:24 CST