PLSQL Performance Issue - sql

Question

This is more of a doubt than a problem. I have a requirement of just selecting some of the common fields from one table and insert it into a different table. I have done the code with 2 different styles but both with BULK COLLECT. Which is the better option to go for or is there any other way apart from this?
Please find the necessary details below.
Procedure (Way 1):
create or replace procedure a2 is
cursor c1 is select id,address from emp1;
type t is table of c1%rowtype;
c t;
begin
open c1;
loop
fetch c1 bulk collect into c;
exit when c.count=0;
forall j in 1..c.count save exceptions
insert into empl1(id,address) values (c(j).id,c(j).address);
end loop;
commit;
exception when others then
for j in 1..sql%bulk_exceptions.count loop
dbms_output.put_line('The sql error is error occured');
end loop;
end a2;
/
Running the above procedure and output:
declare
a number;
begin
dbms_output.put_line ('before procedure: ' || to_char(sysdate, 'HH24:MI:SS'));
a2;
dbms_output.put_line ('after procedure: ' || to_char(sysdate, 'HH24:MI:SS'));
end;
Output :
before procedure: 23:44:48
after procedure: 23:45:47
PL/SQL procedure successfully completed.
So the above procedure took 59 seconds to insert 34801020 records.
Now please find the second procedure.
Procedure (Way 2):
create or replace procedure a3 is
cursor c1 is select id,address from emp1;
type t is table of c1%rowtype;
c t;
begin
select id,address bulk collect into c from emp1;
forall j in 1..c.count save exceptions
insert into empl1(id,address) values (c(j).id,c(j).address);
exception when others then
for j in 1..sql%bulk_exceptions.count loop
dbms_output.put_line('The sql error is error occured');
end loop;
end a3;
/
Running the above procedure with output .
declare
a number;
begin
dbms_output.put_line ('before procedure: ' || to_char(sysdate, 'HH24:MI:SS'));
a3;
dbms_output.put_line ('after procedure: ' || to_char(sysdate, 'HH24:MI:SS'));
end;
Output :
before procedure: 23:47:57
after procedure: 23:48:53
PL/SQL procedure successfully completed.
This Procedure took 56 seconds to insert 34801020 records.
Total records in emp1 table.
SQL> select count(1) from emp1;
COUNT(1)
----------
34801020
Hence My question :
Which of the above two methods is the best way to insert 3 million records into the table and please suggest me if there is any other better way to do the above process of insertion.

Answer 1

I rans a test using a similar piece of code with a similar dataset size
Using a cursor loop
create or replace procedure a2 is
cursor c1 is select empno,ename from bigemp;
type t is table of c1%rowtype;
c t;
begin
open c1;
loop
fetch c1 bulk collect into c;
exit when c.count=0;
forall j in 1..c.count save exceptions
insert into bigemp2(empno,ename) values (c(j).empno,c(j).ename);
end loop;
commit;
exception when others then
for j in 1..sql%bulk_exceptions.count loop
dbms_output.put_line('The sql error is error occured');
end loop;
end a2;
SQL> exec a2
PL/SQL procedure successfully completed.
Elapsed: 00:00:56.93
Doing a regular insert statement, not using a cursor-for loop
SQL> insert into bigemp2( empno, ename )
select empno, ename from bigemp t2
29360128 rows created.
Elapsed: 00:00:11.30
Now do a direct path insert
SQL> insert /*+ append */ into bigemp2( empno, ename )
select empno, ename from bigemp t2
;
29360128 rows created.
Elapsed: 00:00:06.01
Add some parallelism
SQL> alter session enable parallel dml;
Session altered
SQL> insert /*+ append parallel(2 ) */ into bigemp2( empno, ename )
select /* parallel( t2, 2 ) */ empno, ename from bigemp t2
;
29360128 rows created.
Elapsed: 00:00:03.52
So in summary, just by using the appropriate technique, we can make the process go and order of magnitude faster (about 16x faster)