Is EA really scalable?

[Geert Bellekens]

Hi,

Does anyone have experience with scaling EA to a big corporate scale?

I'm talking 2000+ users in 5 different countries. The current models contain something in the order of 10.000.000 (10 million!) objects.

Any feedback is welcome.

[thomas.kilian]

10.000.000 is no number any database would be afraid of. It's more a problem of getting the stuff distributed. There are quite some discussion how to get that to work but it's different in any implementation.

[«Midnight»]

Comma and period are reversed in some regional number schemes.

[Also, if you type like me they appear to be reversed on some keyboards, some of the time...]

[Geert Bellekens]

It's not the database that I'm worried about. They have oracle databases that are far larger then that.
The things that I have most doubts about are:
- the loading speed of the client when trying to open a model with millions of objects
- the problem of accessing the model across 15 sites in 5 countries

[Thomas Mercer-Hursh]

It sounds like you already have the model.  If so, what is your experience with the speed of load with a single user?  If that is acceptable, then the rest is just an ordinary problem of getting a database to perform.

[Geert Bellekens]

Yes the model is there, but the tool is Mega
I would love to sell EA here, but I'm not sure whether EA can cope with a massive setup like this.

PS I don't have a percentage on the license price or something. I'm only hoping to have the pleasure of being able to work with a half decent modeling tool.

[Thomas Mercer-Hursh]

While I haven't focused on the topic, it seems like there have been a number of threads on the forums about very large projects and how to manage them.

I would think the thing to do in this case is to get a trial copy, attempt the transfer, and then try it.  As I say, if it works for one user, the rest is just database management.

[Geert Bellekens]

Not really, the biggest challenge of all will be to get modelers in all those remote locations to work on the same model, with a reasonable performance.
Given the chattiness of EA with the database I don't even want to try to setup a central database to serve all sites.
Possible options would be:
- A cytrix/terminal server type solution (with all the problems associated to that)
- A setup where each location has its own database server keeping a copy of the model. In theory those databases could behave like the setup where each modeler has a local eap file, keeping it synchronized using a VC system (possibly distributed). If I recall correctly that would require a small change from Sparx to create an inbetween solution between the options "This model is private" true/false.

[Jeff Odell]

Assuming you have read through:

http://www.sparxsystems.com/downloads/whitepapers/EA_Deployment.pdf

Depending on your user base and their familiarity with source code control, this might be a good candidate for a source code control approach.

[Geert Bellekens]

I've been emailing with Sparx support, and they have told me that they are working on an option as I described, to have the option "get latest package" when working on a DBMS.
That would indeed allow a setup with multiple database servers locally on the different remote locations, kept in synch using a version control system.
That feature would not be release next build, but maybe the one after that.

[Dave_Bullet]

I hate to say it, but Sparx should fix the product.

The project load time should be similar regardless of whether a project contains 1 package and 1 element or 1000 packages with 100,000 elements.  that = scalability.

It is perfectly possible by doing the following (I have done it on tree based systems). Limit recursive retrieves on any hierarchy to max depth configured by the user.  It's a load vs. navigation performance tradeoff.  There is a setting in the options but it doesn't seem to have any effect.

EA seems to want to inspect every object (as noted in the status bar on a large project open) - when all it needs to do is build the top level of the project browser.  I don't know what it is doing within each element at project load.  Seems redundant to me.

David.

[«Midnight»]

The whole question of lazy load is one that crops up in many products. This is the kind of thing that comes back to 'bite' you later, when the original design was so successful that users scale it well beyond what was imagined - if thought went into capacity planning at all - in early days.

Often this feature falls prey to reluctance to revisit core 'engine' design. The longer it is deferred the greater the (perceived or actual) impact of dealing with it becomes.

A good rule of thumb can be to monitor the degree to which the user community expresses distress with the current design.

Given the extraordinary number of times this has come up, the breadth of contexts where it causes problems to be expressed, and the clearly increasing frequency of complaints, I believe we - and Sparx - have an answer.

Over to you, Sparxians...