Hello, I am currently creating a parallel projectile solver that uses around 8 VMThreads. However, even with that many threads, when solving 10,000 or more projectiles, large stutters occur in the framerate. Currently, the algorithm is as shown: I have placed the hashing process before the actual worker threads because I wanted to see if that would help performance, so that is temporary. Other than that, essentially, this uses a spatial hashing algorithm in order to sort the projectiles to certain workers. I suspect that the hashing of the projectiles is what is causing most of the lag, however, I do believe that the task.spawn for loop is actually causing some lag too. The reason is that the Scheduler is still running a loop per worker thread concurrently.
Any tips for improving the performance?
Just some side information, each worker calculates BOTH the collision and physics step.
WORKER_LOAD = {}
game["Run Service"].Stepped:Connect(function()
debug.profilebegin("hashingProjectiles")
for i, v in next, projectiles do -- ! Fairly optimized, but how do I optimize this further... !
if v.collided() then continue end
local data = v.data() -- Returns a reference, not a copy
local x = math.floor(data.position.X / GRID_UNIFORM)
local y = math.floor(data.position.Y / GRID_UNIFORM)
local z = math.floor(data.position.Z / GRID_UNIFORM)
local h = math.abs(bit32.bxor(x * 92837111, y * 689287499, z * 283923481))
WORKERS[(h % NUM_WORKERS) + 1]:ScheduleWork(data.position, data.velocity, data.id)
WORKER_LOAD[(h % NUM_WORKERS) + 1] = (WORKER_LOAD[(h % NUM_WORKERS) + 1] or 0) + 1
end
debug.profileend()
end)
task.wait(2)
game["Run Service"].Heartbeat:Connect(function() -- Impliment a fixed time step.
debug.profilebegin("workerSpawner")
for _, worker in next, WORKERS do
task.spawn(function()
local work = worker:Work()
if work == nil then return end
for _, v in next, work do
reverse_search[v[3]].updatePos(v[1], v[2])
reverse_search[v[3]].updateCollision(v[4])
reverse_search[v[3]].update()
if reverse_search[v[3]].collided() then -- This can be optimized 100%
--table.remove(projectiles, table.find(projectiles, reverse_search[v[3]])) -- Not performant
reverse_search[v[3]] = nil
end
end
end)
end
WORKER_LOAD = {}
debug.profileend()
end)
