Volker, see my comments below ...

How would you identify an user within your application?

• The user goes on an existing website and creates a new "mobile device" aka mobile user.
• We give him a username and password and at the same time make a call to the SQL Anywhere server, creating a new entry in the ml_users table
• The user logs into the app on his iPad/Android/Win 8.1 tablet and we start the MobiLink sync with these credentials.

Do you use database users, MobiLink user names or the like?

We use the MobiLink user for the authentication on the mobile device.

Do your users get an "empty" database or is it somehow pre-registered (say, with a particular remote_id)?

We deploy an empty UltraLite db with the app

Are your users allowed to move the app between devices (or at least to "unregister" for an old device and to "re-register" for a new one)?

Yes, this is possible. For this, they go to the website, delete the old mobile device/user and create a new one.

I think we have everything in place, what we really would need is "something" when the initial synchronization happens. Then we need to see if the user (aka mobile device) already requested data before. If that is the case, we would return an error message.

I have an idea how this could be done via a small Java program, I just wonder if I really need to do this or if there is a standard functionality that enables this.

Then we need to see if the user (aka mobile device) already requested data before.

Please define "already requested data before". Remote databases upload and download data, they don't "request" anything per se.

return an error message

Please define "return an error message". The MobiLink server processes the upload stream and sends a download stream of inserts, updates and deletes. It does not have a client-server relationship with the remote user, so there is nowhere for it to display anything.

In other words, there is nothing built in to do what you want to do... assuming I understand what you want to do, which I might not :)

If the ML user name is used as "unique user ID" and the database is initially without a remote ID (as you say, it's delivered "empty"), you might choose to use that name as the remote ID, too, as described here.

I guess then you could verify that

a) either the remote id has not yet sync'ed (it has been empty before) b) or it's internal progress state does match the one as stored within the ML server or to compare other "identifying data" as MAC address or whatever.

I guess that can be done with the help of custom authentication (as Vlad has suggested).

Note: Just some suggestions, I have not implemented such a mechanism myself...