The social network for Oracle people

What is your idea to improve INSERT performance?

When we need to insert data(table) to database (NO RAC and RAC) faster...,
we should have disk i/o faster. that's a good idea (example: solid state or make raid 1+0)

INSERT INTO XXX VALUES(a,b,...);

Anyway What should we do on Oracle Database?

- insert nologging
- No Archivelog
- No index
- Use big block size(example 16k, 32k)
- Concern about redo log, if use logging

What about your idea to improve this case?

Tags: insert, table

▶ Reply to This

Replies to This Discussion

Permalink Reply by Faheem latif on February 3, 2009 at 10:52pm

Hi ,

im using "insert /*+ APPEND */" hint option and its good as comapre to others.

Faheem.

▶ Reply

Permalink Reply by Surachart Opun on February 4, 2009 at 12:17am

if i insert with sub query, append hint, that is good.

It's not good with INSERT INTO XXX VALUES (.....,...)
Example:

begin
for x in 1 ..100
loop
insert into t01 values(x, 100);
commit;
end loop;
end;
/

Elapsed: 00:00:00.16

begin
for x in 1 ..100
loop
insert /*+ append */ into t01 values(x, 100);
commit;
end loop;
end;
/

Elapsed: 00:00:01.15

▶ Reply

Permalink Reply by Gerald Venzl on February 4, 2009 at 12:58am

Hi Surachart,

Well there are many answers and my experiences have shown that it always depends on the use-cases.

What you can do to make inserts faster:

1) Bulking (if you want to make it fast, do bulking!) which reduces commits and commit time
2) Enough ITL on the block (INITRANS)
3) Depending on the size of a row: Blocksize, PCTFREE
4) No Indexing
5) No integrity checks (Foreign keys, not null constraints)
6) Buffer cache (if it's to small you'll early end up with I/O)

And the list goes on....

APPEND hint is also very useful on inserts because it's using DIRECT-PATH inserts which will append the data simply on the table. But be aware the unused space by the table will then not be filled up!

Also not forget paritioning!

Regards,

Gerald

▶ Reply

Permalink Reply by Filipe Martins on February 4, 2009 at 1:27am

Hi Surachart,

Gerald made an excellent exposition and I totally agree withg him.
From my personal experience I would recommend partitioning. But my experience is in the telecom world were we could easily make 20/30 million inserts per hour and some of these tables were partitioned by (day,hour) to improve insert speed, we had indexes and we could not use /*+append*/ because of backup restrictions.
If you need real insight or your specific case, may be you could post your full scenario.

Regards,
Filipe

▶ Reply

Permalink Reply by TongucY on February 4, 2009 at 1:54am

I suggest before any guesses first you take an 10046 Level 8 SQL Trace and analyse it with tkprof to see where you spend the time with your inserts.

ps: If there are triggers they also might be a problem area.

▶ Reply

Permalink Reply by Filipe Martins on February 4, 2009 at 3:00am

Hi, do you have a column in the table to store the radius server id (i.ex server1, server2, ...)?
If so, is it indexed? Is it partitioned?
maybe you could consider partitioning the table according to the server. For instance
if table looks like this
c1, c2, radius_server_id, ...

You could use range or list partition by radious_server and the on each server, when you issue the insert statement you could specify the corresponding server id.

create table abc (c1 number,c2 number, radius_server varchar2(14)) partition by list (radius_server) (
PARTITION s1_server VALUES ('s1') ,
PARTITION s2_server VALUES ('s2') )
;

insert into abc partition (s2_server) values(1,2,'s2') ;

▶ Reply

Permalink Reply by Filipe Martins on February 4, 2009 at 3:01am

If the table is indexed by the server_id column, then using partitioning on that table and converting the index to local would produce even better results on insert.
Cheers,
Filipe

▶ Reply

Permalink Reply by Surachart Opun on February 4, 2009 at 3:38am

Thank You for your suggestion. that's a good idea.

Actually this table has kept about account, nas ip, etc... when users log on...
I have 4 indexes. I plan to recreate hash partition, But I think hash can not to improve.

Perhaps I might recreate with range(month) partition.

About list partition, I have no idea to choose about column... now.

Do you have any idea?

▶ Reply

Permalink Reply by Filipe Martins on February 4, 2009 at 3:43am

If your planning to rebuilt table and add partitions, i don't think hash would do the trick on this one...
I stand by adding a range or list partition using a column server_id and forcing each server to specify its own partition while doing the insert.

insert into abc partition (s2_server) values(1,2,'s2') ;