Re: Using a partition as an APPSERVER with RDB

On 2/17/06, Mohammad Tanveer <mtanveer@xxxxxxxxxxxxxxxxx> wrote:
> *         We have 5 partitions on i5.
>
> *         Each partition has many production environments.
>
> *         Each environment is using the same application.
>
> Currently we are maintaining the source for each environment separately.
> Don't ask why (it's been this way for years).  We don't have any problem the
> only issue we have is that when we have to make a global change (very often)
> we have to retrofit this change in 50 different environments (most of the
> time its frustrating), time consuming and requires a lot of resources.

You don't indicate the scale of the "local" differences.  If the
fraction of different code is small, you can automate most of that
away, either with a third-party application or by writing your own. 
Have one version be the master and have programs copy it to the other
versions.  Then have a library in each environments that holds "local"
changes.

> What I am thinking to do is to make one Partition an Application Server (I
> am stealing this term from WEB/Java World), convert the RPGII/RPGIII
> application into SQLRPGLE Application (This will help me to avoid any kind
> of level checking in case there is a slight difference in database in
> individual partition).

The effort involved might be put to use better figuring out how to
localize the differences into a single application, perhaps using
control flags in a location control file.

> Application will run on this partition, depending upon the USER LOGIN,
> application will connect to a particular partition and environment using
> CONNECT database (RDB).  Programs will run on this partition and access the
> remote database. I have done the testing and its working fine.

This might not be necessary, and seems like overkill if you have
localization properly done.

> What I don't know is the performance issues especially when 100s of people
> will connect to this APP server.  Is there any documentation or design
> pattern I must follow so that performance won't become an issue?  Is there
> anyone doing it this way, have they encountered any problem?

Generally, KISS is the usually a good indicator of the best approach. 
I suspect ou are overcomplicating the approach.  Clean up the
application, make it serve all of the locations in a single code base,
and the rest is straight-forward.

I worked at a company that had 14 (slightly) different implementations
of the same system running on a single partition of 1 iSeries, using
the same base code with different "local" library over the base code. 
It worked quite well.

Good luck.

--
Tom Jedrzejewicz
tomjedrz@xxxxxxxxx