Re: What is the best way to do this?

From: Michael Moore <michaeljmoore_at_gmail.com>
Date: Mon, 4 Apr 2011 13:35:15 -0700
Message-ID: <BANLkTikGn0f_L60gpTU3uM1bOB35cgwMdg_at_mail.gmail.com>



Hi Stephan,
 I currently have my Function in a package. The function calls several sub-functions. I've used RESULT_CACHE all all sub-functions where it makes sense to do so. You are correct, it does save a significant number of queries. In my testing, without RESULT_CACHE, each sub-function would be called 30,271 times. WITH Result_Cache, one of the sub-functions was only called 7,126 time and another sub-function was called only 661 times. Quite a savings.

Regards,
Mike

On Mon, Apr 4, 2011 at 1:19 PM, Stephane Faroult <sfaroult_at_roughsea.com>wrote:

> Michael,
>
> What you can possibly do is package your function, and record in a package
> array or whatever, dynamically, attributes that are usually called together
> (if customer_number = customer number of the previous call, then the virtual
> column name probably has some "affinity" with the previous one).
> Whenever you fetch data for a customer, check whether the customer_number
> is the same one as the previous one you have seen, if this is the case
> whether by chance you already hold the value in a buffer. If not, fetch all
> the values that you know to have some affinity with the one that is required
> at once.
> Should avoid a significant number of queries.
>
> SF
>
>
>
> On 04/04/2011 07:20 PM, Michael Moore wrote:
>
> Hi Yechiel,
>
> I can't really do that because I am limited to modification only within my
> PLSQL FUNCTION. I can not change the program that calls my function.
>
>
> My task is to optimize/rewrite an *existing function* which looks like:
>
> FUNCTION get_attribute_value (customer_number IN NUMBER,
> virtual_column_name IN VARCHAR2)
> returns varchar2;
>
> -- where the *return value* is the value of *one* of ATTR001 thru
> ATTR201.
>
> My function does not "know" *all* of the ATTRibute names, it only know
> the ONE that it is currently processing.
>
> There might be, for example, 31, 37 or 121 attributes that have a mapping
> defined, but maybe only 1 of those attributes is required for pricing. For
> example:
>
> abc_key = 778899
> customer # 12345
> attribute07 = 3.5 -- student GPA
> attribute121 = 95050 -- zip code
> attribute129 = Agriculture -- program of interest
> attribute134 = 25 -- age
>
> Now lets assume that for this customer we have the business rule that
> says:
> If the student's age > 23 charge the customer $21
> If the student's age <= 23 charge the customer $29
>
> My function will be passed (778899,'student_age')
> My function will SELECT customer# from ABC where abc_key = 778899;
> My function will figure out that for customer # 12345, 'student_age' is
> stored in attribute134.
> My function will select attribute134 from ABC where abc_key = 778899;
> My function will return '25'
>
> attribute07, 121 and 129 are completely irrelevant to my process as they
> have no affect on the price.
>
> Thanks for your help,
>
> Regards,
> Mike
>
>
>
>
>
>
>
>
>
> On Sun, Apr 3, 2011 at 11:02 PM, Yechiel Adar <adar666_at_inter.net.il>wrote:
>
>> Based on your answers I think I will do this:
>>
>> 1) Create a new table that contain the customer number and the 5
>> attributes that you need for each customer.
>> I mean, do a pivot on the data in data map and create one row for
>> each customer containing all the 5 fields you need.
>> You will need to recreate, or update, this table whenever the data
>> map table changed or when you add or delete a customer.
>> If you have, in data map, old customers that has no data in ABC then
>> you can filter out their mapping in this step.
>>
>> 2) Read the new table. ordered by customer id, and construct dynamic sql
>> to get only the fields you need from ABC.
>>
>> This way, you do not read the big table and you accessed the fact table
>> only once for each customer and you save, big time, on network and database
>> work to put all the fields in your buffer.
>> This method also allow you to use bulk read for data map, saving more
>> time.
>>
>> You did not mentioned what partitioning you have, but if you partition
>> data map according to the field meaning,
>> I mean a partition for all "credit ratings", a partition for all "last
>> order" etc,
>> You can replace the the table in step one with select the will bring you
>> efficiently only the rows you need from the big table.
>> If you have old data map for customers that has no rows in ABC, you can
>> do:
>> select * from data_map
>> where field_description in ('credit rating','last order')
>> and customer_id in (select customer_pk from ABC);
>>
>> Yechiel Adar
>> Israel
>>
>>
>> On 02/04/2011 02:03, Michael Moore wrote:
>>
>> Six questions:
>>
>> 1) How many customers?
>>
>> 26,000
>>
>> 2) Once established, can a customerís data_map change?
>>
>> it's very rare
>>
>> 3) If the answer to #2 is no, or only vary occasionally and with a
>> defined maintenance event to reorganize that customerís data, how many
>> different data_map rows do you have?
>>
>> 1,700,000
>> 5) Do you have partitioning?
>>
>> yes
>>
>>
>

--
http://www.freelists.org/webpage/oracle-l
Received on Mon Apr 04 2011 - 15:35:15 CDT

Original text of this message