And seems I failed to use correct forum account, but anyway, my answer.

If you have not yet looked/used https://www.prosysopc.com/products/opc-ua-simulation-server/, https://www.prosysopc.com/products/opc-ua-browser/, https://www.unified-automation.com/products/development-tools/uaexpert.html plus checked the samples that come with the SDK plus read the tutorials, I would recommend to do so.

In general it all depends what you want to do, but please note that the “solution-space” for OPC UA is basically anything you can do with an object oriented programming/modeling language (i.e. “basically anything”), thus it is hard to condense basics in a single post (typically explaining those takes 3 full workshop days: theory+client+server and that is just to get started). And it also means that there are numerous things that can influence how I will answer depending how the “big picture” looks.

Yes, you can do basically any service call in the client side. On the server side, assuming you use UaNodes to build the address space, Read/Browse will work out-of-the-box. If you do not want to use UaNodes, you will need to do similar construct that the MyBigNodeManager class does in the sampleconsoleserver example, which is basically providing the data when the SDK ask for it. Based on your answer, the MyBigNodeManager way might be easier to implement, but it is basically a bit lower-level API SDK-wise, since most of the advanced features sort of are done on top of the UaNodes currently.

Since interoperability is one of the main points of OPC UA, there is always the question that how a 3rd party client could operate on a Server. If you are doing both the Client and the Server and never ever intend to have any other Client connecting to the Server, you can do things that would not otherwise be recommended (still, in general, it is best to do things the “OPC UA way” where possible). This means you can e.g. do OPC UA Method (see e.g. MyMethodManagerListener in the sampleconsoleserver example) that you would call from the client side to return e.g. a NodeId array per the filter conditions that you would set as input arguments. You can also try implementing the Query like explained in prev post, but it might be a bit too .. complicated for this use-case.

Typically in an OPC UA Server a Variable will be part of a larger modeled Object type that has a lot of metadata. But in this case you could also have the (simple) Variables just inside a simple FolderType instance. Not sure of how many data points you have in total though, it might not be the best of ideas to put e.g. 100k-1million of more references under a single node. Like you can, but a Client doing a Browse on the node might do it some time (e.g. SDK defaults are to ask 1k references at a time per service call). However note that the nodes should still be Browse-able via hierarchical references from the Root/Objects/ FolderType instance (this is one the well known NodeIds a Client can use), but if 3rd party clients are not a thing, then it would not matter (though, then it is more like using OPC UA in a custom solution as an internal implementation detail, not actually “doing” OPC UA so to say).

This is getting somewhat long, so I’ll wait for your answer/opinions before continuing..