Oleg commented once he can make a sim more real (complex) with perfect graphics but thers still no computer in this world which will be able to deal with all that in dogfight without serious performance issues...so optimisation is one of the most important thing in simulation development....

It's very much makes sense when the aim is for high fidelity at every level rather than half-assing it with smoke and mirrors, tricks and cheats, to give the appearance of a battlefield that you have little chance of fully interacting with. It'd just be a side show, because each unit would require a global basic damage and movement model. Boring.

It sounds to me as though Neoqb have not yet got the art of lazy load/unload down. What they apparently do is load everything in memory at mission start.

A relatively simplistic way to do it is ...
1. Large objects (aircraft models) that would cause a noticeable stutter should be pre-loaded.
2. Models of lesser importance (ground objects) should be made with much less detail to avoid taking too much space.
3. Smaller objects/skins/etc. should be lazy loaded (loaded only when called for).
4. Load only what is in view or close to view range.
5. Unload when no longer needed.

The full advantage of on demand loading are not realized due to up front load of large objects, but it would help.

You could even get more sophisticated at some risk.
For example:
Lazy load everything.
Make low detail models for viewing at a distance.
Avoid loading in combat unless absolutely necessary (necessary = view distance is close enough to demand load even though player is in combat. Stutter may occur).
Mitigate the risk of stutter by getting sophisticated about the load algorithm: try to lazy load on a different thread at low priority when player is just flying around. Bump the priority of the load thread if the object is getting closer to the player.

The problem with this is less well defined behavior. It is complex and can be a #%&*$# to debug. There will be "perfect storm" cases (i.e. lots of different aircraft of different types with different skins appearing simultaneously) that could cause severe stutters or memory errors.

The advantage is much better memory usage, usually much better performance, and vastly better initial load time.

Publicly you state the same restrictions to keep people from creating "perfect storm" missions. Under the hood, however, your software is much more capable and load time greatly decreases.

Please take an hour of your time downloading and trying Empire Total War demo. It worths it, even if you don't like strategy games.

What is boring to me is a battlefield with max 40 ground units.

So, haltux is saying "keep the air combat aspect detailed but give the player a simplified, yet greatly expanded, ground combat aspect to immerse himself in the the virtual world".
On the other hand you say "everything should be detailed as much as possible and since we can't do that due to processing restrictions and having the main focus on the air combat, let's not do anything at all".

Is that the correct way to read into it, or did i misunderstand something? Because if i didn't misunderstand it, i think haltux has a better point and his idea would result in a much better virtual environment.

First of all, this is about the online portion and not the single player segment if you'll take note.

So, let's say you have this massively active battlefield of generic pieces that have group mentality and make movements as units rather than individual pieces. What fun is it going to be to take a 2 seater up when they are flyable and attack a ground unit that doesn't take any evasive action? You know you HAVE to bomb that ground unit to accomplish the mission, so thousands of other units out there won't be applicable. I'd rather ground units that take evasive action individually rather than the generic herd mentality in MP.

This is not really a problem. Once one unit requires individual behaviour, because for example it is in your shooting range, it can be broken away from its group and starts behaving according to a complex, individual model. Then, it will come back to its simpler, global, herd behaviour model once it leaves your shooting range.

The number of unit which will requires the complex, individual model will always be very small compare to the overall number of units.

That's just the AI behavior side, but in herd mentallity they have to take obstacles as a group and can't move individually.

The problem with the memory requirements is the detail that goes into the ground units as far as textures go. While some may be fine with boxes moving along the ground that look like they belong in EAW, I prefer detailed ground objects. Looks terrible when the entire world is extremely detailed and then you have a 1997 era style model that you crash next to.

That's already in there. Agree on everything else though. I don't get it why plane models and textures are no preloaded, it causes MASSIVE slowdowns when flying by other planes.

The ground units are already very "reduced", the average texture for vehicles is 256x256 (= about 80 KB), many are 128x128 and the models are also very small (average less than 200 KB). Like i said before, i don't think the numbers presented by NeoQb for ground units are correct, i don't see why ROF couldn't handle more than 40 ground units, it could probably even handle 400 ground units of the same type (talking about memory restrictions only).

If I read this right, they say that you need to reboot after 3 to 4 hours of play. Sorry, but that's a memory leak, and I have no idea why they're trying to blame the OS for that.

And the 2GB of virtual memory limitation shouldn't even come into play. 3 GB of RAM + 2 GB of virtual memory = 5 GB of total memory. If they can't run this sim in that much memory, something is really FUBAR'd.