[topic] Memory footprint

[tuk0z]

Guys, mind adding some of your 0 A.D. games memory footprint here, to help devs get a larger picture while the game is still in Alpha ?

Quite a few around here seem surprised any time someone reports the game's taxing a bit over 1 GB of memory. Some (other I guess ) are working on the subject. Since neither Stan nor I know of a thread already reporting 0 A.D. memory loads from various players / system specs, then here's one.

## Total memory used by 0 A.D. after <TIME> playing on the following system and settings

- Time: 1h08: 4.4 GB ( 4.3 gb RAM + 70 mb swap) - 2 games with same results
- Time : 1h16: 4 GB (all in RAM)
- Time: 1h31:  9 GB (6 gb RAM + 3 gb swap)
- Time: 3h12'. 13.5 GB (7 gb RAM + 6.5 gb swap)

That is with

- Game: a23.1 vanilla (well I added AutoCiv yesterday)
- Map: mainland (small, 300 pop)
- Players: 2 (me + AI very hard balanced)
- Settings: everything On except HQ water effects and vSync which are Off.
- Host: Intel Core i3 @3.2Ghz with 8 GB RAM and a Nvidia GT 1030 2GB.
- OS: Arch linux x86_64

[1] Notes

Atm I took all measures below upon finishing a game while it displays the game stats. I'll *try* to measure at even time such as 30min or 1hour. @Angen and @Stan` I should be able to do the same on dev version as compilation just finished while writing this, finally (/me reading the '10 GB free space' did the trick  ).

Using top and ps which should be available from scratch on any various kind of OS with a in it. More specifically (to get a one shot print)

$ top -b -n1
$ ps aux| grep pyrogenesis

## References

- Optimization is Not Important Right Now? - @sphyrth 2019-10-30
- High memory consumption [2.5 GB] on Windows - @jmauro 2019-06
-

-   [memory] Avoiding system freeze thx to dynamic swap space for Linux players - tuk0z 2019-12  

Edited December 22, 2019 by tuk0z
+ One more reference

[tuk0z]

And here is my first measure with a24.

Total memory used by 0 A.D. after <TIME> playing on the same system and settings as above for a23.1

- Time: ~2h: 2 GB (all RAM) not 100% sure about the exact game duration as overlay time does not work yet.

@Mister the devs I'll give you my impression: after a similar game on a23 about half my storage would be filled with swap and launching eg Firefox-Nightly would freeze the whole system for a good few seconds (note that's on Arch with kernel-ck and muqss ON). With a24 I launched Nightly and pof, the oil factory browser was opened and ready the next second, game still running. Men, the difference is already super impressive At the very least @sphyrth won't have to shred his wallet to buy a gazillion RAM on his next config for sure

Edited December 12, 2019 by tuk0z

[tuk0z]

17 hours ago, Stan` said:

@tuk0z Thanks for investing time into looking into those memory issues I'm wondering if we could pinpoint the commit that fixed the issue for you. Also what do you mean the gametime overlay doesn't work?

Overlay, ie game and real time: I set one or both to ON in settings but none shows up in the a24 game.

Which commit fixed greatly improved the memory usage: Where do we start?

Maybe adding a level of complexity, I'll start by reporting a frank decrease in a23 (twenty three) memory usage while playing long games here, since the time I compiled a24 (Dec 12th). Eg. 2h matches now consistently tax 5-6 GB memory here, leaving my swap alone. Considering the a24 binaries are as independent from the rest of my system as possible, maybe I'd list what updates happened system wise since Dec. 12th:

fmt (6.1.1-1 -> 6.1.2-1), libcap (2.27-2 -> 2.28-1), gconf (3.2.6+11+g07808097-4 -> 3.2.6+11+g07808097-5) and intel-ucode (20191115-2 -> 20191115-3)

Note that on Dec 15th I also updated the kernel, nvidia (440.36-1 -> 440.44-1) and other, but that was yesterday, 2 days after the improvement above appeared (and stayed the same up to now).

EDIT: Options haven't changed at all since the measurements above, with everything On except HQ water effects and vSync which are Off.

 

 

 

Edited December 16, 2019 by tuk0z