Memory consumption and the next official release

[keith.lamothe]

Yea, this is the sort of thing that probably won't actually cause a problem-visible-to-the-player unless it's a really long session or a really large game.  120k+ foreground objects is a lot, though easily achievable with mine enthusiasts and reachable in big/long games.

What I'm interested in from pretty much any significant play-session, though, is the memory consumption of the AIWar.exe at the beginning and end (and preferably some mid-points) and after a re-load of the save at the end, to see if it's holding on to tons of memory.

But thanks very much for the "it's not dying here" note, it is comforting to know that people can still play the latest version

[Fleet]

I'm stressing the system now...I'll report back shortly.

7.3/7.3 Mine Enthusiast and Special Forces Captain
120 Planet snake, with 8 homeworlds.

Main menu start: 100,000K
92405 foreground objects at game start, 610,000K mem after 15sec.
100,200 total ships, 932,000K @ 10min
102,100 total ships, 955,000K @ 15min (end, no more time!)

Restarted, 102,040 FG objects, memory started at 615,000K at start of game reload, rose quickly, and leveled off at 735,000K (I didn't have time to wait and see, but it stopped climbing rapidly, and seemed more stable).

I should be able to get results from a longer game later..

[keith.lamothe]

Thanks for the info.  Been stressing my system too trying to run CLRProfiler on this thing... it assimilated my hard drive!

[vonduus]

I am playing on a 80 planet grid map against an Entrenched Homeworlder and a Sledgehammer, diff 7/7.

Main menu: 100.464 KB

44.485 foreground objects

load: 353.596 KB
15 min.: 529.540 KB
30 min: 519.388 KB
45 min.: 551.720 KB
60 min.: 531.492 KB

I saved the game and restarted: 369.448 KB

[keith.lamothe]

Thanks, that's good info.

Pouring over CLR profiler logs... the difficulty here is that it's something that comes up over a long runtime and the CLR profile would completely fill my hard disk in an hour, tops, and only simulate about 10 minutes of game time in the process

So may have to do more of a hunt-and-peck on all code that's changed since 3.060, but the profiler is helpful in other ways.

[keith.lamothe]

Played for an hour an a half to day (completed the 120 planet map *dances*)

Congratulations

[Rod Serling]

Well, this probably isn't specific enough, but after playing about 3 hours last night, saving, leaving the game paused, and playing about 5 more hours (I'm at in-game hour this morning, I was at 1.2 GB of memory used. Restart, reload the save, and cut down to about 400k

Not sure where to find total foreground objects.

[keith.lamothe]

The info is helpful in that it confirms that the leak is still happening.

The foreground object number comes up during load, and can be seen using in the F3 Debug info as "Total Ships in Game".

Well, I've found a few things that could potentially be contributing to the problem, and fixed them.  Still no smoking gun, though.

[Rod Serling]

Total ships in game was 49,000. Oh this was all 3.0.8.7

[keith.lamothe]

Yea, 49000 is pretty normal for a non-mine-enthusiast midgame.  Thanks

[Atrieus]

I see a significant jump in memory allocation when sentinel frigates are in play.  In particular, when I jump into an AI planet with 200 or more AI sentinel frigates with a largish fleet (750+ ships).  There does not seem to be a reduction when my fleet leaves the system, or is destroyed.

Perhaps the huge range has a target list growing outside of bounds?  Or might have something to do with the modifier to dmg, caused by multiple of the frigates in the same planet?


Atrieus

[vonduus]

3.087

main menu: 95.584
45.035 fg objects

load: 452.324 KB
4 hours: 870.300 KB
6 hours: 904.752 KB
8 hours: 981.148 KB

after restart:

main menu: 95.808 KB
49.343 fg objects

load: 374.004 KB

Unrelated sidenote: The repair tugs have stopped hauling Dyson Gatlings, thanks for the fix.

[Kalzarius]

3.087

Time	Server	FG Objects	Client
Startup	100MB	--	104MB
Load*	361MB	98,410	--
Post-Load	431MB	98,410	390MB
0:15	621MB	97,979	500MB
0:40	697MB	97,652	620MB
1:10	708MB	98,587	625MB
1:40	728MB	99,119	640MB
2:40	750MB	100,056	616MB
3:00	797MB	101,260	631MB
3:30**	980MB	91,480	833MB
4:00**	1070MB	89,163	917MB
5:00**	1097MB	84,155	931MB
5:15**	1124MB	83,155	945MB

* Prior to client connection
** During battle

Although I wasn't paying attention, the client noticed that upon quit, memory consumption jumped to 1.4GB.

[keith.lamothe]

Thanks for the data

Has anyone seen out-of-control memory consumption in 3.091?  I caught a real biggie a couple prereleases ago and between that and other fixes that could have been it.

[Kalzarius]

3.087

Time	Server	FG Objects	Client
Startup	100MB	--	104MB
Load*	361MB	98,410	--
Post-Load	431MB	98,410	390MB
0:15	621MB	97,979	500MB
0:40	697MB	97,652	620MB
1:10	708MB	98,587	625MB
1:40	728MB	99,119	640MB
2:40	750MB	100,056	616MB
3:00	797MB	101,260	631MB
3:30**	980MB	91,480	833MB
4:00**	1070MB	89,163	917MB
5:00**	1097MB	84,155	931MB
5:15**	1124MB	83,155	945MB

* Prior to client connection
** During battle

Although I wasn't paying attention, the client noticed that upon quit, memory consumption jumped to 1.4GB.

For 3.093, the numbers were much more favourable.  FG Objects started at 83,412 and ended at 78,102 with an average server memory usage of 850MB and an  average client usage of 750 MB.  Both server and client jumped about 40-60MB upon application close, at 929MB and 860MB, the peak memory usage of the game.

Things are definitely improved, and memory usage didn't only fluctuate in an upward direction.