Comments on: Pivoting in SQL using the 10g Model Clause

By: Dinakar

Dinakar — Tue, 11 Oct 2011 13:51:16 +0000

Hi Lucas,
i tried using your pivoting logic to convert columns into rows, but i din’t quite understand whats happening in the code though. can you please please explain or answer the below questions:
1.) why a partition was required in this case?
2.) why measures have to start with 0 hrs, why it should be defaulted?
3.) What is the model clause actually doing, can you please elaborate it step by step?

By: Oliver

Oliver — Wed, 30 Nov 2005 17:32:20 +0000

How could I resolved the issue in which I don’t know before hand the number of columns. You know that you have mon, tue … sun. How about if you are trying to convert 20 departments and you want to filter by dept ID but you need the Department Name as the column name. The deparments filtered vary constantly so the names for the columns have to be according to the data extracted

Dept ID Desc Emps
10 Planning 3
10 Planning 3
15 Mktg 2
..
200 Call Center 5

I need an output
Planning Mktg
6 2

another time
Planning Call Center
6 5

By: Lucas

Lucas — Thu, 16 Dec 2004 22:05:09 +0000

From what I recently understood is that this may indicate that there was not enough memory space to hold the intermediate results and file swapping took place for the TEMP area. Can you somehow enlarge the memory available to this process?

By: amihay gonen

amihay gonen — Thu, 16 Dec 2004 20:20:43 +0000

here is a performace test i’ve maded on worked_hours.

insert into work_hours select mod(rownum,56),
rownum,mod(rownum,7),1 from dba_objects;

select count(1) from worked_hours;
49647

select weekno
, empno
, mon
, tue
, wed
, thu
, fri
, sat
, sun
from worked_hours
model
return updated rows
partition by (weekno, empno)
dimension by ( day )
measures ( hours, 0 mon, 0 tue, 0 wed, 0 thu, 0 fri, 0 sat, 0 sun)
RULES upsert
(
mon [0] = hours [1]
, tue [0] = hours [2]
, wed [0] = hours [3]
, thu [0] = hours [4]
, fri [0] = hours [5]
, sat [0] = hours [6]
, sun [0] = hours [7]
)
/

Statistics
———————————————————-
10 recursive calls
0 db block gets
184 consistent gets
1260 physical reads
0 redo size
1228664 bytes sent via SQL*Net to client
36900 bytes received via SQL*Net from client
3310 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
49635 rows processed

49635 rows selected.

Elapsed: 00:00:04.21

SQL> select weekno
2 , empno
3 , mon
4 , tue
5 , wed
6 , thu
7 , fri
8 , sat
9 , sun
10 from (
11 select weekno,empno,sum(decode(day,2,hours,0)) mon,
12 sum(decode(day,3,hours,0)) tue
13 ,sum(decode(day,4,hours,0)) wed
14 ,sum(decode(day,5,hours,0)) thu
15 ,sum(decode(day,6,hours,0)) fri
16 ,sum(decode(day,7,hours,0)) sat
17 ,sum(decode(day,7,hours,0)) sun from
18 worked_hours group by weekno,empno)
19 /

49635 rows selected.

Elapsed: 00:00:01.92

Statistics
———————————————————-
4 recursive calls
6 db block gets
184 consistent gets
272 physical reads
0 redo size
943540 bytes sent via SQL*Net to client
36900 bytes received via SQL*Net from client
3310 SQL*Net roundtrips to/from client
0 sorts (memory)
1 sorts (disk)
49635 rows processed

It seems that the old pivot method is better. It seems that the difference connect somehow to phiscal reads.