How dicrease the time for a bult insert.

Question

Hi, we have a the following process..
1) create a cursor
2) insert in a table t1 acording the rows
selected in the cursor
3) a before insert on t1 create others rows on t2, t3...
This proccess cteate about 300,000 rows a take near 3 hours... my question is how dismiss this time???

Answer 1

Lets imagine I want to INSERT 10 rows into the following table, set a sequential product i.d. for each record and a quantity of zero:
CREATE TABLE sales (product_id NUMBER(10),
quantity NUMBER(5)) ;
Traditionally, I might write something like this (O.K., I know I could use a single INSERT statement for this, but let's assume you want to do some additional calculation inside the loop) :
BEGIN
FOR i IN 1..10 LOOP
INSERT INTO sales (product_id,quantity)
VALUES(i,0) ;
END LOOP ;
COMMIT ;
END ;
All well and good - except each INSERT causes the PL/SQL engine to request that the SQL engine should insert 1 row. This is an overhead that can be reduced by using BULK BIND coding (I mixed my terms previously - a BULK COLLECT allows multiple rows to be retrieved into a PL/SQL collections - it uses the same performance-enhancing technique, but in reverse) :
DECLARE
TYPE t_sales IS TABLE OF sales.product_id%TYPE INDEX BY BINARY_INTEGER ;
a_sales t_sales ;
BEGIN
FOR i IN 1..10 LOOP
a_sales(i) := i ;
END LOOP ;
FORALL j IN a_sales.FIRST..a_sales.LAST
INSERT INTO sales (product_id,quantity)
VALUES (a_sales(j),0) ;
COMMIT ;
END ;
The INSERT above is now passed to the SQL engine as a single request, which gives a significant performance improvement for large numbers of rows. Just remember that all those rows need to go somewhere in memory prior to the INSERT - namely the PGA for your session.
There are a few limitations to this method, the most significant being that fact that the PL/SQL table must be "simple" - i.e. no tables or records.
(Oracle Moderators: Please don't ask "Why is Andy writing such an pointless piece of code?" - it's just to demonstrate the technique!)
null