Questions about performance: Cybran build drones

@rezy-noob said in Questions about performance: Cybran build drones:

i was always wondering why just not making simspeed++ or quality of performance a ranked mod ?
it already solves the problem and doesn't require an official update but whatever,good stuff you put down there Jip,hopefully people will work on it or at least take it in consideration

The simspeed++ mod doesn't actually do a whole lot. It just changes hives in a drastic manner last time I checked.

@valki said in Questions about performance: Cybran build drones:

I think this would significantly alter the look, more than needed... I think you should look at disguising removed sparkle effects - perhaps by one large somewhat random sparkle effect that encompasses the entire build.
Lastly I want to mention that, in 2007, Supreme Commander made me aware of subtractive 3d printing through the Cybran. Preserving that part of the look would be most important to me.

There will be a difference when you use one or two builders. But late game, you'll use dozens of builders. And you won't have the time to appreciate the animation anyhow. Having more performance at that time

If you have 2 - 3 engineers building something then it will still roughly look the same - random drones sending beams and sparkles at different locations. In the end, especially late game when it starts to matter, you typically build things with dozens of builders. And in those specific situations, it will feel the same but it will also be a lot faster and that is what matters.

I got a question from @CheeseBerry about not seeing emitters at all - does it still matter? When building, the answer is yes:

  • 150 T3 t3 Aeon engineers cost about 0.2 ms (!)
  • 150 hyves (upgraded) cost about 11 ms.

A work of art is never finished, merely abandoned

From my understanding you would modify the unit itself. Stupid question amyway, but can this optimisation be done only on a certain amount of engis?

"Nerds have a really complicated relationship with change: Change is awesome when WE'RE the ones doing it. As soon as change is coming from outside of us it becomes untrustworthy and it threatens what we think of is the familiar."
– Benno Rice

@brutus5000 said in Questions about performance: Cybran build drones:

From my understanding you would modify the unit itself. Stupid question amyway, but can this optimisation be done only on a certain amount of engis?

Yes it can be. I am also working with ambient sounds and seeing how that impacts the sim. What you could do is keep track how many of a unit exists (that should be cheap to do) and act on that during unit creation (to prevent unnecessary checks during runtime). For example, you pre-define how many drones a unit should have.

Problem with this is that it will no longer be the same, all the time. And with aesthetics that are visible that may be strange.

edit: with that said, just because we can doesn't mean that I think we should do it. It introduces additional complexity to something that (in my opinion) doesn't require it.

A work of art is never finished, merely abandoned

Because simspeed lag is one of the worst aspects of this game, anything we can do to ameliorate the problem is a great idea. I don't care how many cybran drone bots are visible. In fact, I would probably PREFER to see a laser beam similar to aeon because it makes it a lot easier to see exactly what the hives are assisting. Sometimes you don't notice they started assisting your smd or something else unimportant when that was your lowest priority.

I had the same thought as Brutus. Since 1 hive alone doesn’t cause a huge simspeed problem, let’s leave it as it is and only change the animation once the number of hives hits a critical number where you can’t tell the difference anyways.

@askaholic said in Questions about performance: Cybran build drones:

I had the same thought as Brutus. Since 1 hive alone doesn’t cause a huge simspeed problem, let’s leave it as it is and only change the animation once the number of hives hits a critical number where you can’t tell the difference anyways.

Although one thing to consider is that this will increase the complexity of this portion of the code quite significantly. So we also have to consider if this is worth it from a maintenance stand point.

Im pretty sure in OG FA the number of build bots is exactly as you describe it, 1/2/3 per tech level. ACU has smth like 5?
Then FAF changed it, it was scaled by build power. Then the lag started, so it got nerfed and now with your changes we're going full circle, back to the original values 🙂

@sheikah That question applies to almost all performance optimisations.

"Nerds have a really complicated relationship with change: Change is awesome when WE'RE the ones doing it. As soon as change is coming from outside of us it becomes untrustworthy and it threatens what we think of is the familiar."
– Benno Rice

The key performance issue with Cybran build bots is rather simple - unit count. Every one of those drones is a unit - and they are being created, and destroyed, at a rather fantastic pace. This has two long term side effects - each drone creates a new entity (unit) which has to be initialized just like any other unit - and when it's finished a task - it is destroyed. This process, of itself, is not very hard on the CPU except for the intangible effects of garbage collection.

The garbage collection process of SCFA can, by the time a game is more than 30 minutes old, consume almost 20%, or more, of all the clock cycles available to the SIM. Why ? Because the creation of new units, and new tables and variables in the code are too intensive - and the memory used must be reclaimed - only to be used again - in very quick succession.

The largest of all memory allocations in SCFA is the creation of a new unit. Anyone who has ever examined a blueprint for any unit knows just how many variables are involved - and that is the problem in a nutshell. If you reduce the creation of drones down to a modest level, you minimize this problem quite a bit. Now - in the end - it becomes a simple matter of aesthetic versus performance.

FAF is reusing the bots, as long as there is another target to build right away, they dont despawn.

Is changing the speed that drones change target from .2s to .5s purely a visual thing or is that a nerf to how quickly they can change what they’re assisting or building?

Rookie game engine question here: how did you evaluate the increase in time to render ? What console commands and such?

I would like to just for fun have some people I know with “cheese or potato” PCs run it, along with my own, see what happens.

Overall though, as Sprouto said, it’s almost as simple a fact that more units are generated.

@exselsior said in Questions about performance: Cybran build drones:

Is changing the speed that drones change target from .2s to .5s purely a visual thing or is that a nerf to how quickly they can change what they’re assisting or building?

Purely visual. It is how often they choose a random location within the building they are 'building'. The resources switches immediately. The beams of the engineer would follow right away. The drones would follow shortly after.

A work of art is never finished, merely abandoned

@Morax use ren_ShowNetWorkStats in the console. Spawn in 150 hives and make them assist a quantum gateway while you have essentially no income.

A work of art is never finished, merely abandoned

@jip Figured that would be the case, thanks! I like the ideas

In order to push the conversation to a conclusion I'd like to sit down with people. I'd like to take out potential ambiguity. I assume that the people who upvoted the original post are positive to the changes.

I'll be on the FAF discord channel on Thursday 21:00 GMT+2 and on Friday 21:00 GMT+2 - please be there if you have questions. I intent to implement the changes this weekend - that way it can be inspected on the Dev branch starting next week and hopefully be live when September starts.

A work of art is never finished, merely abandoned

How hard would it be to make a SIM mod that does this? So we could test it for ourselves. We could test for performance or look at how it affects aesthetics so we can decide evaluate the trade-offs involved.

Also I assume that you would have to basically do the same work to prepare the SIM mod as to make these changes to the main branch of FAF itself. So there is essentially zero wasted effort in taking the time to make this SIM mod.

If you just plug the changes directly into the dev branch, it would be a lot harder to compare with/without the changes, because there may be other differences in the dev branch. And every time we load up a game with the dev branch, doesn't it have to download lots of files? Like if we wanted to go back and forth between regular FAF and the dev branch to test it, it would be a pain. But if you just make a SIM mod it would be easier for us to see the difference.

Alternatively: make a SIM mod that adds new units: a new hive with the changed animations. Then we could run a single game, cheat in some units/buildings, and put regular hives on one side and new hives on the other side. And then we could watch them work at the same time. So we could do a side-by-side splitscreen comparison of the aesthetics with both running. And then run only one side at a time to check performance with the other side hives paused.

The problem is time. I can spent all week on fixing this one performance issue and find something that we can all agree on, make a sample version, etc - or, I can spent a day on it and spent the other six days on other issues. There are hundreds of things that need to be looked at from a performance perspective - this is just one of them 😞 .

If I understood correct then due to recent changes only the changed files need to be downloaded. Therefore there is no large download anymore. But don't quote me on that, as it may not be 100% stable yet.

If you want to compare it side-by-side, launch the game with the FAF Develop version and then directly launch the executable of FAF found in the bin folder (Program Data/ForgedAllianceForever/Bin). The latter should use the default init file where as the former is running FAF develop. Then you can view the changes side-by-side.

edit: the purpose of this topic is to make sure the aesthetical change (which is, again - minor in my opinion) is accepted before I do the work. As I do not want to see my time go to the bin.

A work of art is never finished, merely abandoned

@arma473 if you enable caching in the client settings the download will only need to happen once. Also I think it won't need to happen multiple times without that because they are actually different files. But turn it on just to be safe.

We limit the number of unique targets. Instead of having each drone (and the engineer build arms) all have their own unique target, they all share the same target (as the engineers). With that we limit the number of entities (to 1 / builder) and the number of emitters (just 1 sparkle emitter / builder).

Do I understand this correctly, that this would make all the drones do their laser "welding" at the same spot?