Advice on materialized view.

Question

Hello everybody,
I have a question and I hope you can advise. Imagine the following (simplified) model.
CREATE TABLE players
  player_id NUMBER(15) PRIMARY KEY,
  player_name VARCHAR2(100) NOT NULL,
  creation_date DATE,
  birth_date DATE
  -- around 20 other fields (firstname, different sizes, weights...)
CREATE TABLE competitions
  id_competition NUMBER(15) PRIMARY KEY,
  competition_name VARCHAR2(100),
  start_date DATE,
  end_date DATE,
  creator_id NUMBER(15), -- FOREIGN KEY TO PLAYERS.
  winner_id NUMBER(15)   -- FOREIGN KEY TO PLAYERS.
  -- around 20 other fields
CREATE TABLE objectives
  id_objective NUMBER(15) PRIMARY KEY,
  objective_name VARCHAR2(100),
  id_competition NUMBER(15),  -- FOREIGN KEY TO COMPETITIONS
  creator_id NUMBER(15),   -- FOREIGN KEY TO PLAYERS.
  winner_id NUMBER(15),    -- FOREIGN KEY TO PLAYERS.
  start_date DATE,
  is_public NUMBER(1),
  restrictions VARCHAR2(100)
  -- around 20 other fields
);[Foreign keys are all indexed]
The table players has around 30 million records.
The table competitions has around 10 million records.
The table objectives has around 70 million records.
I have a materialized view which JOIN between all these tables. Something like this:
SELECT comp.id_competition, comp.competition_name, /* comp.... */
       comp_cre.player_name, /*, comp_cre....*/
       comp_win.player_name, /*, comp_cre....*/
       obj.objective_name,   /*, obj....*/
       obj_cre.player_name,  /*, obj_cre....*/
       obj_win.player_name   /*, obj_cre....*/
       -- some analytics rank() OVER, COUNT() OVER...
  FROM competitions comp
  LEFT JOIN players comp_cre
         ON comp.creator_id = comp_cre.player_id
  LEFT JOIN players comp_win
         ON comp.winner_id = comp_win.player_id
INNER JOIN objectives obj
         ON obj.id_competition = comp.id_competition
  LEFT JOIN players obj_cre
         ON comp.creator_id = obj_cre.player_id
  LEFT JOIN players obj_win
         ON comp.winner_id = obj_win .player_id;THe materialized view is refreshed everyday. But since we have huge amount of data it takes a while to run. The size of the materialized view is more than 15GB. It is normal that the plan generate full table scans on these 3 tables... Since, that the goal. REtrieve all the data to make a big result set;
I was wondering if any of you has an idea on how I can speed up the refresh? I can not add materialized view logs on each dependent table. Is it possible to tell Oracle to work in different Thread? SOmething like the first 5 milions competitions are done in a thread, and another 5 milions in another thread? I have a very robust machines. 128GB ram and 64 CPU. Any suggestion?? I'm using Oracle 10g.
THank you

Answer 1

With 64 CPUs, you should certainly look at parallelizing the MV refresh:
http://www.doug.org/newsletter/march/MV_Refresh_Parallel.pdf
Iordan Iotzov
http://iiotzov.wordpress.com/