Help to improve the performance of a procedure.

Question

Hello everybody,
First to introduce myself. My name is Ivan and I recently started learning SQL and PL/SQL. So don't go hard on me. :)
Now let's jump to the problem. What we have there is a table (big one, but we'll need only a few fields) with some information about calls. It is called table1. There is also another one, absolutely the same structure, which is empty and we have to transfer the records from the first one.
The shorter calls (less than 30 minutes) have segmentID = 'C1'.
The longer calls (more than 30 minutes) are recorded as more than one record (1 for every 30 minutes). The first record (first 30 minutes of the call) has segmentID = 'C21'. It is the first so we have only one of these for every different call. Then we have the next (middle) parts of the call, which have segmentID = 'C22'. We can have more than 1 middle part and again the maximum minutes in each is 30 minutes. Then we have the last part (again max 30 minutes) with segmentID = 'C23'. As with the first one we can have only one last part.
So far, so good. Now we need to insert these call records into the second table. The C1 are easy - one record = one call. But the partial ones we need to combine so they become one whole call. This means that we have to take one of the first parts (C21), find if there is a middle part (C22) with the same calling/called numbers and with 30 minutes difference in date/time, then search again if there is another C22 and so on. And last we have to search for the last part of the call (C23). In the course of these searches we sum the duration of each part so we can have the duration of the whole call at the end. Then we are ready to insert it in the new table as a single record, just with new duration.
But here comes the problem with my code... The table has A LOT of records and this solution, despite the fact that it works (at least in the tests I've made so far), it's REALLY slow.
As I said I'm new to PL/SQL and I know that this solution is really newbish, but I can't find another way of doing this.
So I decided to come here and ask you for some tips on how to improve the performance of this.
I think you are getting confused already, so I'm just going to put some comments in the code.
I know it's not a procedure as it stands now, but it will be once I create a better code. I don't think it matters for now.
DECLARE
CURSOR cur_c21 IS
    select * from table1
    where segmentID = 'C21'
    order by start_date_of_call;     // in start_date_of_call is located the beginning of a specific part of the call. It's date format.
CURSOR cur_c22 IS
    select * from table1
    where segmentID = 'C22'
    order by start_date_of_call;
CURSOR cur_c22_2 IS
    select * from table1
    where segmentID = 'C22'
    order by start_date_of_call;  
cursor cur_c23 is
    select * from table1
    where segmentID = 'C23'
    order by start_date_of_call;
v_temp_rec_c22 cur_c22%ROWTYPE;
v_dur table1.duration%TYPE;           // using this for storage of the duration of the call. It's number.
BEGIN
insert into table2
select * from table1 where segmentID = 'C1';     // inserting the calls which are less than 30 minutes long
-- and here starts the mess
FOR rec_c21 IN cur_c21 LOOP        // taking the first part of the call
   v_dur := rec_c21.duration;      // recording it's duration
   FOR rec_c22 IN cur_c22 LOOP     // starting to check if there is a middle part for the call
      IF rec_c22.callingnumber = rec_c21.callingnumber AND rec_c22.callednumber = rec_c21.callednumber AND 
        (rec_c22.start_date_of_call - rec_c21.start_date_of_call) = (1/48)                
/* if the numbers are the same and the date difference is 30 minutes then we have a middle part and we start searching for the next middle. */
      THEN
         v_dur := v_dur + rec_c22.duration;     // updating the new duration
         v_temp_rec_c22:=rec_c22;               // recording the current record in another variable because I use it for the next check
         FOR rec_c22_2 in cur_c22_2 LOOP
            IF rec_c22_2.callingnumber = v_temp_rec_c22.callingnumber AND rec_c22_2.callednumber = v_temp_rec_c22.callednumber AND 
              (rec_c22_2.start_date_of_call - v_temp_rec_c22.start_date_of_call) = (1/48)        
/* logic is the same as before but comparing with the last value in v_temp...
And because the data in the cursors is ordered by date in ascending order it's easy to search for another middle parts. */
            THEN
               v_dur:=v_dur + rec_c22_2.duration;
               v_temp_rec_c22:=rec_c22_2;
            END IF;
         END LOOP;                     
      END IF;
      EXIT WHEN rec_c22.callingnumber = rec_c21.callingnumber AND rec_c22.callednumber = rec_c21.callednumber AND 
               (rec_c22.start_date_of_call - rec_c21.start_date_of_call) = (1/48);       
/* exiting the loop if we have at least one middle part.
(I couldn't find if there is a way to write this more clean, like exit when (the above if is true) */
   END LOOP;
   FOR rec_c23 IN cur_c23 LOOP             
      IF (rec_c23.callingnumber = rec_c21.callingnumber AND rec_c23.callednumber = rec_c21.callednumber AND
         (rec_c23.start_date_of_call - rec_c21.start_date_of_call) = (1/48)) OR v_dur != rec_c21.duration          
/* we should always have one last part, so we need this check.
If we don't have the "v_dur != rec_c21.duration" part it will execute the code inside only if we don't have middle parts
(yes we can have these situations in calls longer than 30 and less than 60 minutes). */
      THEN
         v_dur:=v_dur + rec_c23.duration;
         rec_c21.duration:=v_dur;               // updating the duration
         rec_c21.segmentID :='C1';
         INSERT INTO table2 VALUES rec_c21;     // inserting the whole call in table2
      END IF;
      EXIT WHEN (rec_c23.callingnumber = rec_c21.callingnumber AND rec_c23.callednumber = rec_c21.callednumber AND
                (rec_c23.start_date_of_call - rec_c21.start_date_of_call) = (1/48)) OR v_dur != rec_c21.duration;                 
                // exit the loop when the last part has been found.
   END LOOP;
END LOOP;
END;I'm using Oracle 11g and version 1.5.5 of SQL Developer.
It's my first post here so hope this is the right sub-forum.
I tried to explain everything as deep as possible (sorry if it's too long) and I kinda think that the code got somehow hard to read with all these comments. If you want I can remove them.
I know I'm still missing a lot of knowledge so every help is really appreciated.
Thank you very much in advance!

Answer 1

Atiel wrote:
Thanks for the suggestion but the thing is that segmentID must stay the same for all. The data in this field is just to tell us if this is a record of complete call (C1) or a partial record of a call(C21, C22, C23). So in table2 as every record will be a complete call the segmentID must be C1 for all.Well that's not a problem. You just hard code 'C1' instead of applying the row number as I was doing:
SQL> ed
Wrote file afiedt.buf
  1  select 'C1' as segmentid
  2        ,start_date_of_call, duration, callingnumber, callednumber
  3  from (
  4        select distinct
  5               min(start_date_of_call) over (partition by callingnumber, callednumber) as start_date_of_call
  6              ,sum(duration) over (partition by callingnumber, callednumber) as duration
  7              ,callingnumber
  8              ,callednumber
  9        from table1
10*      )
SQL> /
SEGMENTID  START_DATE_OF_CALL     DURATION CALLINGNUMBER   CALLEDNUMBER
C1         11-MAY-2012 12:13:10 8020557824 1982032041      0631432831624
C1         15-MAR-2012 09:07:26  269352960 5581790386      0113496771567
C1         31-JUL-2012 23:20:23  134676480 4799842978      0813391427349
Another thing is that, as I said above, the actual table has 120 fields. Do I have to list them all manually if I use something similar?If that's what you need, then yes you would have to list them. You only get data if you tell it you want it. ;)
Of course if you are taking the start_date_of_call, callingnumber and callednumber as the 'key' to the record, then you could join the results of the above back to the original table1 and pull out the rest of the columns that way...
SQL> select * from table1;
SEGMENTID  START_DATE_OF_CALL     DURATION CALLINGNUMBER   CALLEDNUMBER          COL1       COL2       COL3
C1         31-JUL-2012 23:20:23  134676480 4799842978      0813391427349          556         40       5.32
C21        15-MAR-2012 09:07:26  134676480 5581790386      0113496771567          219        100      10.16
C23        11-MAY-2012 09:37:26  134676480 5581790386      0113496771567          321         73       2.71
C21        11-MAY-2012 12:13:10 3892379648 1982032041      0631432831624          959         80       2.87
C22        11-MAY-2012 12:43:10 3892379648 1982032041      0631432831624          375         57       8.91
C22        11-MAY-2012 13:13:10  117899264 1982032041      0631432831624          778         27       1.42
C23        11-MAY-2012 13:43:10  117899264 1982032041      0631432831624          308         97       3.26
7 rows selected.
SQL> ed
Wrote file afiedt.buf
  1  with t2 as (
  2  select 'C1' as segmentid
  3        ,start_date_of_call, duration, callingnumber, callednumber
  4  from (
  5        select distinct
  6               min(start_date_of_call) over (partition by callingnumber, callednumber) as start_date_of_call
  7              ,sum(duration) over (partition by callingnumber, callednumber) as duration
  8              ,callingnumber
  9              ,callednumber
10        from table1
11       )
12  )
13  --
14  select t2.segmentid, t2.start_date_of_call, t2.duration, t2.callingnumber, t2.callednumber
15        ,t1.col1, t1.col2, t1.col3
16  from   t2
17         join table1 t1 on (   t1.start_date_of_call = t2.start_date_of_call
18                           and t1.callingnumber = t2.callingnumber
19                           and t1.callednumber = t2.callednumber
20*                          )
SQL> /
SEGMENTID  START_DATE_OF_CALL     DURATION CALLINGNUMBER   CALLEDNUMBER          COL1       COL2       COL3
C1         11-MAY-2012 12:13:10 8020557824 1982032041      0631432831624          959         80       2.87
C1         15-MAR-2012 09:07:26  269352960 5581790386      0113496771567          219        100      10.16
C1         31-JUL-2012 23:20:23  134676480 4799842978      0813391427349          556         40       5.32
SQL>Of course this is pulling back the additional columns for the record that matches the start_date_of_call for that calling/called number pair, so if the values differed from row to row within the calling/called number pair you may need to aggregate those (take the minimum/maximum etc. as required) as part of the first query. If the values are known to be the same across all records in the group then you can just pick them up from the join to the original table as I coded in the above example (only in my example the data was different across all rows).