Multicore FA

[SC-Account]

I wondered how difficult it would be to add multi core support to FA. Is it possible to get access to the core files? And how much of it is Lua code?

[rootbeer23]

you need the source code for the game to implement multithreading.
which you dont have.
then you need to implement a multithreaded design which produces exactly the same results on all clients.
which is very complicated.
it is already a hard problem to make it work with a single thread.
an educated guess is that it takes 1 person several months of work.

[Kof]

Never going to happen IMO, multi-threaded programming is much more complex and difficult to implement. I'd estimate it would take man years to re-develop.

[rootbeer23]

one way to do it is to draw an imaginary thick cross on the map (imagine the red cross flag), which gives you 4 white quadratic regions and everything that happens in one white square has no chance of being affected by anything in another white square (with some exceptions) during a single tick in the simulations.
what you do then is launch 4 threads, 1 per white square and simulate the 4 squares.
note that what is inside the 4 white squares can be affected by entities in the red cross... that is why the thread that
processes the red cross is inactive while the white squares are simulated and run afterwards.
the runtime of one round of simulation is then the time it takes to simulate 1 white square (4 will run in parallel and finish at approximately the same time) plus the time for the red cross.
but the execution order will be perfectly synchronized.

[Benoker]

In this model an ASF blop fight can happen in one square, and all the multicore does not help because this one square is only handled by one core...

Personally I think it does not matter in which order the collisions between projectiles and units are calculated because total damage is a simple addition of all the single damages (a+b = b+a). Each core could randomly choose which projectiles/units to handle.

But if it would be so easy GPG would have implemented it There really is no point in trying to modify the application without even having access to the source code.

[rootbeer23]

In this model an ASF blop fight can happen in one square, and all the multicore does not help because this one square is only handled by one core...

nothing is perfect

Personally I think it does not matter in which order the collisions between projectiles and units are calculated because total damage is a simple addition of all the single damages (a+b = b+a). Each core could randomly choose which projectiles/units to handle.

computer1 simulates projectile P34324 hitting tank t232 and destroying it first and computer2 simulates tank232 shooting at tank545 and afterwards simulates projectile P34324 hitting t232.

[Benoker]

Concurrency is nothing that can be solved in one day. But for example the very real issue you bring up might be sovled with "All computers first simulate tanks firing their shots, then all computer simulate tanks taking damage, then all computers calculate which tanks have taken too much damage and explode" -> meaning breaking down the algorithm in several steps, each of which can be solved concurrently - but it would not be possible to execute more than one (big) step at a time. Of course this solution might have some side effects causing trouble elsewhere...

... and just our friendly argument should prove to SC-Account that the answer to "how difficult would it be ..." is "very".

[rootbeer23]

All computers first simulate tanks firing their shots, then all computer simulate tanks taking damage, then all computers calculate which tanks have taken too much damage and explode

you would see a similar problem with just the movement of units which would not be solved by the solution above.

[discoverer2k4]

seperate air, land, engy-like and naval on 4 different threads
somebody allready posted a link to the engine, I thought actually...?! But it would be illegal to change something in the engine without permission of the licence owner...

[SC-Account]

Well, my point was more about getting the code etc. The multi threading itself should not be the problem. I am confident that there is always something you can put in another thread, even if it might not be perfect multi threading it will be better than the current.