o01eg wrote: ↑Tue Nov 10, 2020 8:14 pm
Currently server uses:
Code: Select all
VIRT RES SHR S %CPU %MEM TIME+ COMMAND SWAP CODE DATA USED
842628 60716 13220 S 0.0 12.3 16:49.73 freeoriond 112740 3572 220544 173456
Do you have VIRT values for 12th game (at start and end)?
Anyways, testing locally with single player mode:
freeorion before starting a game (no freeoriond): 1300k
1000 systems, 6 AIs + 1 human:
freeorion turn 1 1320k +/- 10k, turn 100 1380k, turn 150 1410k, turn 200 1450k... looks slightly superlinear, takes 60k for first 100 turns, 70k for next 100 turns, so maybe 1630k for turn 400.
freeoriond turn 1 1100k, turn 100 1100k, turn 200 1100k, with spikes to 1700k (from turn 1).
100 systems and 6 AIs I got worse results:
freeorion turn 1 1330k, turn 100 1460k, turn 200 1530... so for this game we got a much faster growth at start and slower later. We can't expect steady, regular curves here.
freeoriond turn 1 1050k, turn 100 1050k, turn 200 1100k, with spikes to 1650k (from turn 1). Here it seems number of systems takes a small toll on server memory usage: 50k
10 systems and 1 AI:
freeoriond turn 1 1035k, turn 100 1040k, turn 200 1045k... Didn't see any spike, but that might be due to luck (top refreshes once per second, spikes could be in between).
So it seems seed (luck) is quite relevant for client resource consumption, probably due to AIs building ships faster or slower, having more or less monsters, etc. But number of systems doesn't seem to affect resource consumption much, for neither client or server: from 1035k (spikes maybe somewhere around 1600k) with 10 systems to 1100k (1700k spikes) with 1000 systems.
Server only has freeoriond, right?
o01eg wrote: ↑Tue Nov 10, 2020 8:14 pm
I suppose next game could be tried with 38-40 systems per player.
100, please! With research and production cost factors at 50% of default.
If in the end it's a bad idea, you'd find out at game test and cancel game if it takes too much memory?
Imagine that memory usage grows as slow as in my tests above, then by making increases of +10 systems per player we would need more than a year to reach 100 systems per player and find out if it was doable or not.
Edit: regarding game pace (turns to finish), galaxy size isn't as determinant as it seems (for multiplayer), because we humans know how to outpost to quickly reach our enemies and wreak havoc. I mean, doubling galaxy size doesn't doubles time to first contact, time to first battle, etc. If a game with 30 systems per player can be decided around turn 100 (and linger until 150), a game with 60 systems per player could be decided around turn 200 (and finished around turn 250).