Query rewrite without a group by

Question

The actual query is more complicated involving 6 table joins. I have tried to re-produce the issue using a simple example shown below.
The "contacts" table has about 14 columns including the ones shown below.
All the table statistics are up to date.
select * from contacts;
customer_num          media_type          phone_area_code          phone_number
12345               FAXW               322               987-7865
12345               PHH               564               984-9872
12345               PHH               NULL               NULL
12345               PHW               873               922-9291
12345               PHW               762               872-9291
87272               PHH               456               982-9388
87272               FAXW               728               918-9281
76288               PHW               872               982-2882
The criteria is a customer can have more than 1 work or home phone.
work phone is identified as media_type = "PHW".
home phone is identified as media_type = "PHH".The example query is:
SELECT customer_num,
       <col-1>
       <col-10>,  
        MAX(decode(media_type,
                           'PHH',
                           phone_area_code || '''' || phone_number,
                           NULL)) home_phone,
       MAX(decode(media_type,
                           'PHW',
                           phone_area_code || '''' || phone_number,
                           NULL)) work_phone
FROM
   contacts
GROUP BY
      customer_num
      <col-1>,
      <col-10>;The explain plan is good. all indexes are used and the cost is less. The CPU Wait time is high.
How can this query be re-written without a GROUP BY?.For 2000 customers I am grouping 2000*11 records.
This is taking time.
I cannot post the explain plan or tkprof as I am not at work at present.
Can this query be re-written without using MAX(DECODE and GROUP BY?
Thanks in advance!

Answer 1

user1980 wrote:
How can this query be re-written without a GROUP BY?.For 2000 customers I am grouping 2000*11 records.
This is taking time.You could self-join the table, using a full outer join between the home phone records and the work phone records. I'm not convinced that it would be much faster than the aggregation, and might even be slower, but a nested loop join between the data sets would probably bring back the first rows faster than your current method.
The most efficient self join would probably come from partitioning the records based on the media type so that PHH and PHW records were in their own partitions. A hash join between the two sets would then be cheap. However, it might be worth looking at loading the table into a hash cluster (on customer_num) or using an index-organised table (on customer_num and media_type), either of which might improve join speed and also be more sympathetic to better performance in the rest of your application when you (presumably) want to retrieve all records for a particular customer_num.