Fast insert 5000 rows from memory

Question

Hi,
I have a user loading excel data on his screen - its via a webserver application. So now he has a couple of 1000 rows in the memory.
What would be a fast way of inserting this data in a table?
insert (a,b,c,d) into target_table
*10,000
- pl/sql block with insert statements?: not good. pl/sql blocks are limited in length.
- seperate insert statements?
- a procedure that does an insert with bind variables? proc_insert (a,b,c,d)?
It would be great if you can help me out.

Answer 1

S11 wrote:
The client can submit is as structured data, such as seperate insert statements.
What other way could he deliver this in to be faster then 5000 seperate inserts....5000 separate insert statements are NOT slow.
Observe - firing 5000 insert statements using PL/SQL:
SQL> create table foo_tab(
  2          id      number,
  3          d       date,
  4          n       number
  5  );
Table created.
SQL>
SQL> declare
  2          n       number;
  3  begin
  4          n := dbms_utility.get_time;
  5 
  6          for i in 1..5000
  7          loop
  8                  insert into foo_tab values( i, sysdate, dbms_utility.get_time );
  9          end loop;
10 
11          dbms_output.put_line( 'Total insert process: '||(to_char(dbms_utility.get_time-n)/100)||' seconds' );
12  end;
13  /
Total insert process: .88 seconds
PL/SQL procedure successfully completed.
SQL>
SQL> select
  2          (max(n)-min(n))/100             as TOTAL_SQL_SEC,
  3          count(*)/((max(n)-min(n))/100)  as ROWS_PER_SEC
  4  from foo_tab;
TOTAL_SQL_SEC ROWS_PER_SEC
          .82   6097.56098
SQL> During the time this loop was run, the database was basically able to deal with over 6000 inserts per second (from this single session).
Sure, inserts from an external client will be slower. And due to no fault of the database. The client needs to ship the data across network infrastructure to the client. The most effective way of doing this is via bulk processing and packing as much data as possible per TCP packet (as governed by the MTU/Max Transmission Unit size of the network architecture used).
But looking at the pure database side (which is effectively what the above sample code does), the server is able to deal with 5000 separate inserts in less than one second.
So what is the most effective database interface? SQL. Plain and simple.
How do you use SQL itself effectively? Besides coding optimal SQL, it is coding sharable SQL. The above example uses a single SQL cursor in the SQL Shared Pool to execute that 5000 insert statements.
This is not rocket science. ;-)