Re: Database Access Path Question

[Vern Hamberg <vhamberg@xxxxxxxxxxxxxxxxxxxxxxxxx>]

IBM recommends DYNSLT if the number of records selected is over 80%. Here's 
the scoop from InfoCenter. Note the bit on access path sharing, but also 
note the use of "can" in many places - classic "it depends" stuff here.

Dynamic select/omit

With dynamic select/omit, when a program reads records from the file, the 
system only returns those records that meet the select/omit values. That 
is, the actual select/omit processing is done when records are read by a 
program, rather than when the records are added or changed. However, the 
keyed sequence access path contains all the keys, not just keys from 
selected records. Access paths using dynamic select/omit allow more access 
path sharing, which can improve performance. For more information about 
access path sharing, see Using existing access paths.
To specify dynamic select/omit, use the dynamic selection (DYNSLT) keyword. 
With dynamic select/omit, key fields are not required.
If you have a file that is updated frequently and read infrequently, you 
may not need to update the access path for select/omit purposes until your 
program reads the file. In this case, dynamic select/omit might be the 
correct choice. The following example helps describe this.
You use a code field (A=active, I=inactive), which is changed infrequently, 
to select/omit records. Your program processes the active records and the 
majority (over 80%) of the records are active. It can be more efficient to 
use DYNSLT to dynamically select records at processing time rather than 
perform access path maintenance when the code field is changed.

HTH

Vern

At 08:05 PM 2/6/2003 +0000, you wrote:
> Correct.
>
> This is really one of those "it all depends". By using the READE approach 
> you are tackling the requirement through an application program - which 
> means that the same solution has to be "coded" in every program that would 
> require that selection path.
>
> By using the DYNSLT approach, it is being done by the database manager ans 
> all an application (RPG, SQL, Java etc.) has to do is open the file and 
> process it.
>
> As Steve pointed out, the use of DYNSLT can be quiet an overhead depending 
> on the number of row returned.
>
> FWIW, I would try the DYNSLT approach first ('cos I always prefer to do as 
> much as I can in the database as opposed to within a program) and then, if 
> I wasn't happy with the performance, go for the READE.
>
> HTH
>
> Paul