Ze_PilOt wrote:
There is clearly an issue with AIs slowing down the sim as the time goes by.

Some myth says it's because it's still managing wrecks, but it's probably not true, or not entirely true, as even a nuke doesn't help.

Sheeo wrote:
Sorry for the late reply -- I'm busy with uni work

Resin_Smoker wrote:Pilot...

This is how to test a mod: http://www.faforever.com/forums/viewtopic.php?f=45&t=7232&p=71412#p71412

Resin

Well all right, why haven't you posted your scores?

I get ~1.700 with the mod on, and about the same with it off.

I think it'd be better to track actual memory allocations directly, using the profiling tools I've noted before.

But it does look like we'll need to look in other directions to improve simspeed.

Ze_PilOt wrote:
If wrecks have any impact on the sim, a sandbox game with a big amount of wrecks would have a decreased sim speed.
No need to have any other test than that. We are testing wrecks, they are easy to produce in any number.

So far, I'm unable to find any correlation between sim speed & wrecks.

Same for engineers really. People thought they were lagging the game. But engy assisting a factory doesn't impact the sim speed at all. I've proven that assisting doesn't have any impact on sim speed (or very little).
Because they are not moving, like wrecks.
What is impacting sim speed is engy failing to go to the factory or units trying to go through a bunch of engy.

Like wrecks. But as wrecks can't move and units can go through...

What I'm willing to believe, but still need proof, is that wrecks from an AI are impacting the sim speed. I've seen that while testing AI for the latest patch : For every AI dead, the sim speed wasn't increasing at all.
So something was still be computed or not released correctly. Was it wreck? No idea. Maybe, maybe not. Need tests.

In conclusion, while it's good that someone is trying to solve sim speed issues, this is not a trivial change.
It can even impact the sim speed more than it solves it.
So before jumping to conclusion and integrate a code that solve a probably non-existing issue, I would like proofs.

Resin_Smoker wrote:

Sheeo wrote:2. I'm concerned that the overhead of a ForkThread call per wreck shadows the potential performance gain of removing wrecks in the long run. I haven't yet tried running the profiling tools (There are some goodies in /lua/system/profile.lua), but I'll try and do some tests when I have time.

Code:

         -- if wreck is in water, some of his mass is removed.
         if self:GetCurrentLayer() == 'Water' then
            mass = mass * 0.5
            energy = energy * 0.5
         end

The above snippet is from FAF Unit.lua, line 1669... I'm assuming this is what you were referring to?

I've added this to my copy of the mod and will be available on the next update.

Resin

Resin_Smoker wrote:
Also keep in mind that the games Default wreckage script doesn't have a Trash-Bag to clean up after a wreck has been removed from play.
This means that every wreck created is a potential memory leak waiting to happen.

Ze_PilOt wrote:

Resin_Smoker wrote:Just spawning units and letting them sit will not lag the sim much at all if the units are not active.

How am I supposed to move wrecks?

I think that we have first to confirm the fact that wrecks impact the sim speed before thinking on working on something like your mod.

So far, I can't confirm that wreck impact the sim speed in any way.

Testing with and without your mod in a real game won't work, they are way too many other variables. Testing wrecks, and wrecks only, is the only way to see if it's a fact or a myth.

ie. For a long time, people though that ASF were lagging the game.
In reality, they impact the sim less than any other unit, even in really big numbers.
What is impacting the sim is the projectiles they are firing. (really high fire rate for a late game unit).

Resin_Smoker wrote:
Just spawning units and letting them sit will not lag the sim much at all if the units are not active.

IceDreamer wrote:

Ze_PilOt wrote:What is the proof that wreck impact sim speed?

Other than pathfinding, I can't find any evidence of that.

In a sandbox, 3000 wrecks, still running at +10. (while 1000 units alive run at +8)

I've not tried it, but the games where this might make a difference are going to be with far more units and wrecks in play. We're talking late-game Setons, 800 units per player (6400 total) plus three times that many in wrecks. That's a LOT of stuff.