Re: Thinking out loud about a new RPG web framework

[Thorbjørn Ravn Andersen <thunderaxiom@xxxxxxxxx>]

Joe Pluta skrev  den 06-06-2008 15:15:
> Thorbjørn Ravn Andersen wrote:
>   
> > By having a connection for every user session you may run into a 
> > scalability problem since you require the user to log out to regain the 
> > resources allocated to that connection.
> >   
> >     
> What "resources"?  It's a connection to a QZRCSRVS job, limited only by 
> the number of TCP/IP ports.  You put a decent timeout on the session and 
> it works fine.  Please, give me a real world example of a situation you 
> think would not scale.
>
>   
We are talking about running Java on the AS/400, yes?

So the number of AS400 objects in Java, number of TCP/IP connections and 
QZRCRVS jobs scale with the number of users logged in and not the number 
of users active.  There might be a bit of a difference for heavily 
loaded systems.

What is the number of simultaneous users we are talking about here?


> Pooling doesn't help at all with the maximum number of users, unless you 
> are stateless.
>
>   
Sure it does.  If the statefull logic is on the Java side and the 
statelessness is in the RPG code, there is quite a bit to get there.

Your RPG code must not expect things to come in a certain order unless 
you are absolutely certain that users do not jump in the middle of a 
sequence by reposting a form they bookmarked, or just press reload on a 
timed out session, etc.


> > This naturally requires that the backend calls are stateless - all 
> > needed information is passed from the session object - which you must do 
> > anyway since this is for web users.
> >
> >   
> >     
> Some applications can be stateless, but the best performance comes from 
> stateful applications.  If you're using the browser as a replacement for 
> 5250 applications, then you need stateful connections, otherwise 
> performance is unacceptable.
>
>   
Naturally.  I was expecting that we were looking at writing new stuff 
here, instead of gluing a webfront end on code not written to be web 
callable.

> I don't want to go into a long discussion on this particular point.  If 
> you believe that all applications must be stateless, then we have a 
> fundamental disagreement.  Most business application developers I talk 
> to agree that some applications require statefulness, and when I'm 
> talking about a persistent connection, it's always for that class of 
> applications.
>   
As written above, i do not.  I just say that for the java+RPG 
combination I believe that best performance requires that you put the 
session state on ONE side of the gap that the QZRSVRS connection 
imposes.   For java using session beans, this means having all the 
session state in Java.

To me :)