10-40 Player Game (Online)

Feel free to join up, only requirement to see this dedicated server game is to have PA Stats installed

Server Specs

4770k Overclocked to 4 ghz
16 gb of ram (running at 1800)
Normal hard drive

16 GB of ram is still not quite enough for a game this size @cola_colin had a 17 player game on that died half way through because his RAM was filled up.

Well the server had some other pastats related things running.
But yeah the 32 player game went up to 25GB of ram in use.

what about using your harddrive as additional ram? does that help?

That'd kill sim fps.

me understand this do not .... ???

IS IT STILL UP O_O

I'd love to join

It would not crash, but lag HARD. Depending on the type of RAM it can transfer a 5-20 GB of data per second.
A normal HDD can't even do much more than 100 MB/s.

Pagefiling is a looot slower than traditional RAM.

Even SSDs only manage about 500 MB/s or so. Think it'd be enough to eat into the read/write cycles? Or have they mitigated that problem, I couldn't find anything about it.

Join up guys, we need moar people!

It's definitely not healthy, but from some random tests I read a while ago the testers didn't really manage to burn through SSDs even with programs that constantly wrote data to them at max speed for as long as the testers had time, which was long.

Was fun while we lasted, thanks for hosting this.

Personally I prefer no AIs in my game, but I'll join all the big games that I can at this point.

So I guess it really is only an industrial application concern. Seems like they'd be able to help take some of the load, even if its not GB/s

is this why the lobbies are taking so long to fill?

so in other words i couldn´t just plug in a external terrabite harddrive and use half of it expecting much of an increase?

Well the game would not crash. But it would work at like super slow speed.
Basically to generate super sized record "lol so big" planets people have set 100gb or more of HDD as virtual memory before. By doing that they could generate HUGE planets that dont fit in 32gb RAM, but it took DAYS to build a single planet. You would basically rate performance in frames per hour or frames per day.

It's worth trying the sandbox biome for big games. I don't know how much ram a big system takes up on the server, but it's probably a fair chunk. Could make games last longer before hitting the ram ceiling.

Depends if the ram usage growth during the game is related to the system or not.

The first rule about storage is:
The faster you can read/write to it, the more expensive per Byte it get's. You can make the values better by using Volatile memory (power loss = data loss), but that only works for very rare cases.
L1 Cache of the CPU let's average RAM look like a snail. But it is just ways too expensive to get any decent amounts of storage.
Swapping only helps you if you do not need that data anymore (so it getting swapped out on the disk is not an issue as you don't need to access it). Swapping is the worst thing that can happen to data really. Especilyl data you are working on.
Perhaps the best idea would be to use DRAM-based SSD for swapfile. Wich are basically just DRAM in a SSD casing (data loss on power loss; no write cycle limits).
Overall NAND SSD seem to be able to take quite some writeload nowadays. A bigger issue seems to be power loss right now.

As far as I can tell from my personal test the server is currently not recycling memory and seems to access no longer valid data. So it will jsut keep on piling up data in the RAM that it at as least looks at. Loosing AI or players does not reduced the memory profile in any way.
The only time memory is recycled seems to be on application closure. Memory Perforamnce wise it seems to behave like a Garbe Collecion approach where there is never a GC cycle before the end of the programm run.
That does fit with the server only running "one instance per game".
Data will jsut keep piling up as units get produced and never be released as they get destroyed. It could even be that the memory for projectiles is never recycled eitehr.

The other thing is that doing Garbage Collection in a Multithreaded environment is really, really hard. Deterministic memory control is better, but also the source of about 95% of all bugs in programms that use them. There are tons of reasons this appraoch is nowadays frowned upon.
So they have the chocie to either risk random performance losses/stalls. Or opening a worm can full of problems.