Desyncing between host and client without d/c 14.7

[Krizs]

Hello devs,

We originally started this factory out in 12.12 and have spent roughly 2x240 hours with my friend trying to achieve our megabase which can sustain 1 rocket launch / minute while not working with buffers for about an hour.

Unfortunately, we started to get problems somewhere in 14 with the client not being in sync with the host, as in the client having vastly different ups than the host, leading to the client not being able to control his character. It can easily happen that the client keeps on running for about a minute after not giving any input anymore.

We noticed from the debug view show time used percent, that the vast majority of time is spent on trains and entity updates.

We also noticed that the train pathfinding makes it impossible to build rail segments for one of us while the other would be doing something else, as when a rail is placed the pathfinding takes up too much resource to recalculate all train paths and the game becomes very unresponsive as long as we are building trains.

We would like your input on whether we could salvage this game in multiplayer or whether we should just wait for 15.0 with the new optimisations.

Here is the current version of our save on 14.16
https://www.dropbox.com/s/ceuvj6yf6mrwi ... 6.zip?dl=0

Thanks in advance.
Krizs & bloodsparrow

[bk5115545]

The hardest part of pathfinding AI is when a decision has to be made; "does the train turn here or here or here or ...".

I'm downloading the save and I'll check it out tonight (I love looking at megabases) but for immediate improvement you can have less signals and less junctions.


Also, question for the devs here; Do the trains use a pathfinding cache like the biters do? If not then there's an "easy" major optimization.

[Rseding91]

The hardest part of pathfinding AI is when a decision has to be made; "does the train turn here or here or here or ...".

I'm downloading the save and I'll check it out tonight (I love looking at megabases) but for immediate improvement you can have less signals and less junctions.


Also, question for the devs here; Do the trains use a pathfinding cache like the biters do? If not then there's an "easy" major optimization.

Biter path finding vs train path finding are different. Biter paths don't change frequently and trains do. When a train moves into some set of rails it changes the cost for pathing. If a train is stopped on some rail set and a train can path around it it has to run that logic when the time comes and so on.

[bk5115545]

Thanks Reseding. That makes sense.
Here's the debug timings. I have yet to find the massive amount of entities that need updating. Most of the furnaces don't have any ore, there are very few robots in use, and most of the miners are inactive.

Could the number of beacons be the cause of the bottleneck?
EDIT:
It's probably how many signals there are. There looks to be thousands.

[bk5115545]

Turning the base on. Petroleum isn't finished yet but the iron and copper products can be made.

[Krizs]

Thanks for looking into it. As for the factory we have run the oil products module test but the pumps and oil rail network still has to be made as well as the actual launch site. The coal mines also need to be reintegrated into the upgraded 8line railway bus if I remember correctly.The satellite production module is also missing as of now. Then we would need to do the full integration test looking for building errors and then we would hopefully be done. What we do know is that we can produce the required plates and oil products as well as green circuits.