What can I do to improve performance?

Question

CREATE TABLE temp_table
( ts timestamp
, ttl NUMBER(5,0)
, ipid NUMBER(10,0)
, length NUMBER(5,0)
, tid NUMBER(5,0)
, cnxid NUMBER(10,0)
, reverse NUMBER(1,0)
, srcIp VARCHAR2(15)
, srcPort NUMBER(10,0)
, dstIp VARCHAR2(15)
, dstPort NUMBER(10,0)
, flags varchar2(10)
, startSeq NUMBER(19,0)
, endSeq NUMBER(19,0)
, nbBytes NUMBER(5,0)
, ack NUMBER(19,0)
, win NUMBER(10,0)
, urgent NUMBER(10,0)
, options varchar2(500)
CREATE TABLE mf1 NOLOGGING AS SELECT DISTINCT TS, TTL, IPID, LENGTH, TID, CNXID, REVERSE, FLAGS, STARTSEQ, ENDSEQ,NBBYTES, ACK, WIN, URGENT, OPTIONS from temp_table order by cnxid
CREATE INDEX mf1_cid_idx ON mf1 (cnxid) NOLOGGING
CREATE INDEX mf1_cid_ts_idx ON mf1 (cnxid,ts) NOLOGGING
ANALYZE TABLE mf1 ESTIMATE STATISTICS SAMPLE 5 PERCENT FOR ALL INDEXED COLUMNS
This is the function I would like to achieve better performance on, or rather it is the query in cursor c_query. I have a larger application that does a lot more, but we have found that the most time consuming part is doing the c_query on various cnxid so I figured a function like this would be a decent test to see how it performs without having to run the entire application.
CREATE OR REPLACE FUNCTION test1 RETURN NUMBER AS
CURSOR all_cid IS SELECT distinct cnxid FROM mf1;
all_cid_row all_cid%ROWTYPE;
sum_val NUMBER(19,0);
CURSOR c_query(in_cid IN NUMBER) IS SELECT /*+ INDEX(mf1 mf1_cid_ts_idx)*/ * FROM mf1 WHERE cnxid = in_cid ORDER BY ts;
c_query_row c_query%ROWTYPE;
BEGIN
sum_val := 0;
FOR all_cid_row IN all_cid LOOP
FOR c_query_row IN c_query(all_cid_row.cnxid) LOOP
sum_val := sum_val+1 ;
END LOOP;
END LOOP;
RETURN sum_val;
END;
I added a hint in the c_query for which index it should use. Without the hint it use mf1_cid_idx and performance is better with mf1_cid_idx.
Runnig the SQL tuning advisory from enterprise manager I am told to analyse the table, which I though I did and to get another CPU.
I have tried changing the analyse to this without any effect:
begin dbms_stats.gather_table_stats(ownname => 'MARIUS', tabname => 'MF1', estimate_percent => DBMS_STATS.AUTO_SAMPLE_SIZE, method_opt=>'FOR ALL COLUMNS SIZE AUTO'); end;
I have tried clustering, but the way I insert tuples to the mf1 table achieve much the same as long as no new tuples are inserted. In other words when I tried clustering it I got the same performance.
Another thing I have been wondering is that if I populate the mf1 table with 10MB 100MB 1GB 5GB the amount of time per MB of data decreases to a low point at 5GB and then dramatically increase if I populate the mf1 table with 10GB.
The tablespace containing mf1 is 44GB large it only contains one incarnation of the mf1 table. It has a maximum 13GB used when I populate the mf1 with a 10GB table. temp is 8GB, undotbs1 8GB
Any suggestion how I can imporve my c_query?

Answer 1

I''m looking at the bulk transfer, actually I'm running tests to see how it perform. I was wondering if some one that was familiar with BULK transfeer could have a look at my new test1 function to see if it is equivalent to the old one. (Added the old one to the bottom for ease of reference).
CREATE OR REPLACE FUNCTION test1 RETURN NUMBER AS
TYPE number_table is TABLE OF mf1.cnxid%TYPE INDEX BY BINARY_INTEGER;
cid_list NUMBER_TABLE;
TYPE tab_table IS TABLE OF mf1%ROWTYPE INDEX BY BINARY_INTEGER;
tuple_list TAB_TABLE;
sum_val NUMBER(19,0);
i NUMBER;
ii NUMBER;
BEGIN
sum_val := 0;
SELECT DISTINCT cnxid
BULK COLLECT
INTO cid_list
FROM mf1
ORDER BY 1;
FOR i IN cid_list.FIRST .. cid_list.LAST LOOP
SELECT /*+ INDEX(mf1 mf1_cid_ts_idx)*/ *
BULK COLLECT
INTO tuple_list
FROM mf1
WHERE cnxid = cid_list(i)
ORDER BY ts;
FOR ii IN tuple_list.FIRST .. tuple_list.LAST LOOP
sum_val := sum_val+1 ;
END LOOP;
END LOOP;
RETURN sum_val;
END;
CREATE OR REPLACE FUNCTION test1 RETURN NUMBER AS
CURSOR all_cid IS SELECT distinct cnxid FROM mf1;
all_cid_row all_cid%ROWTYPE;
sum_val NUMBER(19,0);
CURSOR c_query(in_cid IN NUMBER) IS SELECT /*+ INDEX(mf1 mf1_cid_ts_idx)*/ * FROM mf1 WHERE cnxid = in_cid ORDER BY ts;
c_query_row c_query%ROWTYPE;
BEGIN
sum_val := 0;
FOR all_cid_row IN all_cid LOOP
FOR c_query_row IN c_query(all_cid_row.cnxid) LOOP
sum_val := sum_val+1 ;
END LOOP;
END LOOP;
RETURN sum_val;
END;
--- Edit, I get the following error trying this.
Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - Production
With the Partitioning, OLAP and Data Mining options
begin
ERROR at line 1:
ORA-04030: out of process memory when trying to allocate 16396 bytes (koh-kghu
call ,pmuccst: adt/record)
ORA-06512: at "MARIUS.TEST1", line 23
ORA-06512: at line 2
This is in the inner query. Do I have to reset or empty tuple_list between each iteration? How is this done?
Message was edited by:
Marius