Simulation speed changes for unkown reason

[Julio Jerez]

how many units are we talking about?
also starting with newton 2.16 is is possible to set newton world to run on separate thread and still run detrministic.
I beleieve you use a lock step model for your world, or it is a server.
but maybe tshi could be useful for running tha extra worlds you have running simultaneswwtly.

is your game simulation run in one server? so that you can use SSE.
also in the graph what does each trace mean.

teh first thing I would do is try to intgrate tow 2.16 and see if nothing break, 2.10 is too old of a version to debug anything for me.
then we can check what trick with can do to speed things up.

[Marc]

How many units? Hmm good question. I have to add some debug stats like that. I initially planned that every player has maybe about 30 units. Now they all have hundreds. And they like - but it's getting quite slow :/

I use a lock step model for my game. I could use SSE - cpu's that don't have that are most likely too slow anyways to run my game. Unless that kills determinism.

The deterministic multithreading features sounds interesting. So you seperate the scene into several groups you have to solve and then you solve them deterministically in parallel? If that's the case, that could help.

Yeah when I start optimizing again, I'll first upgrade to the newest version and check if it works correctly.

I remember you suggested two changes that might or might not help which I haven't tried yet: One was to group all trees in a compound collision and the other was related to the space axis because my ground floor is xy and newton's broadphase is optimized for xz?

I'll check the amount of units there tomorrow as it's already late over here. I'm glad you are so helpful all the time

[Julio Jerez]

I remember you suggested two changes that might or might not help which I haven't tried yet: One was to group all trees in a compound collision and the other was related to the space axis because my ground floor is xy and newton's broadphase is optimized for xz?

I remenber that too those could be big help, but first let us start with teh simple stuff.

- add debug code to twht we can see teh total number of unints.
- color the debug grapch so the we know what is what and whish parth of teh engien are takin teh time.
- update to the latest version of newton 2.16 (many bug fixes and optimizations)


Once we have that infomnation we cna proceed with other thing.
SSE is a huge perfermace jump, jum almos twice as fact speed up.
since you are no using a central cever to do the simulation then mutreading may not help,
but we can try I beleive that setting newton to use internal should generate the same result, if it does not detremonistic maybe we can make it so I do no see why it could not be.

burt even we can get all those optimizations, the gain will not be better than say 2 or 3 hundred % and I do not see teh world so crowd,
this means we need to really kknow what part of teh system is cause the slow down and solve with some software solution that reduces and algorthm complexity
rather than trying to make every thing faster marginally.

[Marc]

Because of the not-anymore-fitting-topic, I continue this thread here: http://newtondynamics.com/forum/viewtopic.php?f=9&t=5637&p=40769#p40769.