Hi
For design a non-trivial distributed database application,
to be implemented in Application server.
Here some guidelines and ideas I picked during my searching, I maybe wrong :)
* Don�t expose entity beans to the clients:
It is recommended that the client will talk with session beans and the
session beans will talk with the entity beans.
This way the client request for operation form the session beans and the
session bean do all the work, by access to multiple entity beans and other
session beans and work as a facade,
hiding from the client the server object schema complexity
and collect the operations a client need to do to one operation
those avoiding network traffic.
* Bean managed persistence is BAD:
Also there is a clear and easy framework how to write the code for BMP.
It is BAD to use it if performance is important.
Using BMP will be VERY inefficient and will generate many small SQL operations.
If you need operation like
SELECT FNAME,LNAME FROM EMP
BMP will generate:
SELECT PKEY FROM EMP
And then for every record in the table
SELECT FNAME,LNAME,ADDRESS,... FROM EMP WHERE PKEY=?
* Component managed persistence is problematic:
Also theoretically the container can do some optimization to overcome
the BMP problem, there still a problems There is no standard how to
specify mapping to the component (Fields, and Find code)
Most of the implementation are trivial and does not allow complex mapping
between beans and database.
Those limitations will be improved in the future, as technology
mature, but it will certainly take time.
But even if CMP is perfect, why we need them?
If the client does not talk with them, They only accessed internally to the server.
Entity beans are heavy they required stub, skeleton, home, component side
object the bean itself and more,
I may be wrong here in the list of objects required but I think
Entity Beans are still heavy compared to simple java classes.
So:
Does entity beans are useful (either now or in the future) for real world database
application?
Maybe what is required is a different standard that fit better to map Java
objects (without the remote features) to databases
something like TopLink, JavaBlend or Cocobase?
Meanwhile is there some design patterns that can be used to design today
persistence objects that can provide some acceptable performance with
current technology and are easy to migrate once a better one come?
Any other ideas how to implement domain logic as objects?
Regards,
Muly.
===========================================================================
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff EJB-INTEREST". For general help, send email to
[EMAIL PROTECTED] and include in the body of the message "help".
