Poor application design

If you have only a poorly defined set of symptoms that might indicate a loop, or a wait, or possibly a performance problem with an individual transaction, consider the possibility that poor design might be to blame.

An example of poor application design is given here, to show how this can give rise to symptoms which were at first thought to indicate a loop.

Environment:
CICS and DL/I using secondary indexes. The programmer had made changes to the application to provide better function.
Symptoms:
The transaction ran and completed successfully, but response was erratic and seemed to deteriorate as the month passed. Towards the end of the month, the transaction was suspected of looping and was canceled. No other evidence of looping could be found, except that statistics showed a high number of I/Os.
Explanation:
The programmer had modified the program to allow the user to compare on the last name of a record instead of the personnel number, which it had done in the past. The database was the type that grew through the month as activity was processed against it.

It was discovered that in making the change, the program was no longer comparing on a field that was part of the key for the secondary index. This meant that instead of searching the index for the key and then going directly for the record, every record in the file had to be read and the field compared. The structure of the source program had not changed significantly; the number of database calls from the program was the same, but the number of I/Os grew from a few to many thousands at the end of the month.

Note that these symptoms might equally well have pointed to a performance problem, although performance problems are usually due to poorly tuned or overloaded systems, and affect more than just one transaction. Performance problems tend to have system wide effects.



dfhs10j.html | Timestamp icon Last updated: Monday, 5 June 2017