JDBC adapter - polling application - is this efficient ?

Question

Hi,
I have a scenario that is requesting data via the JDBC adapter from an SQL server system and sending the resultset into BW.
This is polling every 10 minutes with the following statement :
select * from itecs_dummy.dbo.action
where current_timestamp between
convert(datetime, (convert(varchar(13), GETUTCDATE()) + '23:40:00.000')) and
convert(datetime, (convert(varchar(13), GETUTCDATE()) + '23:50:00.000'))
This basically will return zero records 99% of the time as it check the system time is between 23:40 and 23:50
Eventually I will have around 30 JDBC communication channels performing the same frequency of polling but read data from different tables (the system time check will still be the same for each select)
Is this highly inefficent in XI terms ?
Would it not be better to write a simple ABAP to request the data from BW to the JDBC adapter at a fixed time each day therefore cutting out the unnecessary polling ?
Kind regards
Colin.

Answer 1

Hi Colin,
The method you're using is really the only simple solution to workaround XI's lack of a scheduler.
We have about 10 JDBC interfaces live for a year at a large XI client using this workaround. They're polling every 5 minutes and we've never had an issue.
We decided to use this method based on the following:
1) The load on the DB wouldn't be significant. Even if you have a complex DB select statement, the DB optimizes the call and should run fast if the 'where' clause fails. You can check this with your DBA.
2) The load on XI would be greater if we implemented this in another way. Any other solution would likely require BPM which is much more intensive for XI then these JDBC calls.
We chose 5 minutes instead of 10 because of the following scenario which could occur although it's not likely:
- XI polls at 23:41, message is sent
- XI J2EE goes down at 23:42 and restarts
- XI J2EE back up at 23:47
- XI polls at 23:47, message is sent
So in this scenario, you could get 2 messages sent. Since our J2EE stack takes longer than 5 minutes to restart, we resolved this issue by using a 5 minute polling interval.
I think the more elegant and precise way is to make a call from ABAP, but it would be more complex, and put more load on XI due to the use of BPM. If you take into account the above items I think your current approach is the best.
Thanks,
Jesse