1
Suggestions and Requests / Re: What EA needs for large organizations?
« on: May 24, 2007, 06:13:17 am »
If the performance issues can be addressed then the usability issue may not be that painful. If you need things integrated then you will end up with one large logical model regardless of how you do it.
It is not news that a fat client architecture does not scale. If Sparx wants to seriously crack the enterprise market it will have to move to a multi-tiered architecture. This has HUGE implications and will require a near rewrite of EA, but it has to be done.
In the mean time, rather than loading the complete project at startup, much can be done on an event-driven as-needed basis. Varying Hierarchy View Depth does not seem have an influence on performance. The software should assume the responsibility of providing a user only the data that she needs (or indirectly require) rather than requiring the user to segment the model purely for performance reasons. Changing the performance model to provide many small delays rather than one big delay may be slightly annoying but will definitely scale much better.
Our enterprise model takes upwards of 5 minutes to load and performance in general can be poor at times (I am assuming that you are prepared to accept that the size of the model is the issue and that external Network and Database performance issues have been addressed).
By the way, my name is Chris Kriel. I don't know why the forum software thinks I am Javier Guerra Diez.
It is not news that a fat client architecture does not scale. If Sparx wants to seriously crack the enterprise market it will have to move to a multi-tiered architecture. This has HUGE implications and will require a near rewrite of EA, but it has to be done.
In the mean time, rather than loading the complete project at startup, much can be done on an event-driven as-needed basis. Varying Hierarchy View Depth does not seem have an influence on performance. The software should assume the responsibility of providing a user only the data that she needs (or indirectly require) rather than requiring the user to segment the model purely for performance reasons. Changing the performance model to provide many small delays rather than one big delay may be slightly annoying but will definitely scale much better.
Our enterprise model takes upwards of 5 minutes to load and performance in general can be poor at times (I am assuming that you are prepared to accept that the size of the model is the issue and that external Network and Database performance issues have been addressed).
By the way, my name is Chris Kriel. I don't know why the forum software thinks I am Javier Guerra Diez.