| View Issue Details [ Jump to Notes ] | [ Issue History ] [ Print ] | ||||||||
| ID | Project | Category | View Status | Date Submitted | Last Update | ||||
|---|---|---|---|---|---|---|---|---|---|
| 0003660 | Spring engine | General | public | 2013-03-29 14:40 | 2013-04-23 20:33 | ||||
| Reporter | bibim_ | ||||||||
| Assigned To | Kloot | ||||||||
| Priority | low | Severity | minor | Reproducibility | always | ||||
| Status | resolved | Resolution | fixed | ||||||
| Product Version | 94.1 | ||||||||
| Target Version | Fixed in Version | ||||||||
| Summary | 0003660: Spring 94 + HealthBars widget causing big memory usage increase | ||||||||
| Description | Test system: CPU: AthlonXP 3000+ OS: Windows NT 5.00.2195 SP4 RAM: 2 GB Video: GeForce 7950GT 512MB Test replays: Size: 4v4 (2 E323 + 2 KAIK) vs (2 E323 + 2 KAIK) Map: Tangerine Game: BA 7.72 Duration: replays stopped after 10 minutes in game (~ 200 units) Results for spring.exe process: Spring 91.0, HealthBars enabled: Peak mem usage: 376MB Page faults: 185 354 Spring 91.0, HealthBars disabled: Peak mem usage: 374MB Page faults: 178 966 Spring 94.1, HealthBars enabled: Peak mem usage: 550MB Page faults: 1 448 351 Spring 94.1, healthBars disabled: Peak mem usage: 381 MB Page faults: 199 942 I stopped after 10 minutes each time in order to be able to do the tests quickly, as it's enough to notice the memory increase even if it doesn't seem that big. However, with real life multiplayer games with more units, Spring takes all available memory and my system swaps so much that the game becomes unplayable after about 15 minutes. This can be solved either by disabling HealthBars, either by switching back to Spring 91. I've set the issue severity to minor and priority to low because I didn't find any evidence of reproducibility on other systems than mine. | ||||||||
| Tags | No tags attached. | ||||||||
| Checked infolog.txt for Errors | |||||||||
| Attached Files |
| ||||||||
 Relationships |
|
 Relationships |
| Notes |
|
|
bibim_ (reporter) 2013-03-29 19:22 Last edited: 2013-03-29 19:36 |
I did some additional tests on a different system (Win7 / i5 / 4GB). On this system, enabling HealthBars widget with Spring 94.1 / BA 7.75 didn't increase peak memory usage that much (better Win7 memory management?). However, following results show a really big increase in page faults number when HealthBars widget is enabled: With HealthBars: Initial page faults (just when Spring finishes loading): 195 215 Page faults after 3 minutes in game (normal speed): 2 283 778 => page faults triggered during 3 minutes of game at normal speed: 2 283 778 - 195 215 = 2 088 562 Without HealthBars: Initial page faults (just when Spring finishes loading): 194 974 Page faults after 3 minutes in game (normal speed): 241 832 => Page faults triggered during 3 minutes of game at normal speed: 241 832 - 194 974 = 46 858 So, during the first 3 minutes of this replay, I got 44 times more page faults for spring process just by enabling HealthBars. These results were obtained using following replay: http://planetspads.free.fr/spads/tmp/20130328_214620_Tangerine_94.sdf (I tested several times to avoid side effects, and could reproduce it 100%). Weirdly enough, the problem isn't triggered by all replays. For some replays (even with bigger number of players and/or units), I have very similar page faults numbers with or without HealthBars. Could it be related to the map? (reclaimable features?) Because I noticed the big increase in page faults rate is observed from the start, even when no unit is even spawned yet... |
|
2013-03-29 19:55 |
http://code.google.com/p/zero-k/source/detail?r=9790 maybe relevant? |
|
jK (developer) 2013-03-31 01:03 |
test? |
|
abma (administrator) 2013-04-03 20:30 Last edited: 2013-04-03 20:30 |
@bibim & knorke: can you try with the latest develop build as it shows lua mem usage in debug view? this should help to find the cause... |
|
bibim_ (reporter) 2013-04-04 15:43 |
Same test as in issue description, but this time I played it at 1x speed (which seems to make things worse), with current develop (94.1.1-105-g4169a3d): Peak mem usage: 1.36GB Page faults: 7 117 916 (once again, these values are for about 200 units only...) I had HealthBars enabled during the 10 minutes of the test, and lua allocated memory was continuously changing very fast between ~100MB and ~860MB (maybe more, because I didn't keep an eye on it during the whole 10 minutes => maybe a peak usage line would be useful...). Then after these 10 minutes I disabled HealthBars, and lua allocated memory stayed between 5MB and 6.5MB. |
|
bibim_ (reporter) 2013-04-16 16:32 |
Is there anything planned concerning this, or is it an expected behavior for Spring 94 to have 44 times more memory pagefaults for Spring process when HealthBars is enabled? |
|
Kloot (developer) 2013-04-16 19:20 |
you did mark it low priority... |
|
bibim_ (reporter) 2013-04-16 20:13 |
Yes I did, because initially I couldn't find any evidence of reproducibility on other systems than my old system, and also because jk told me in #sy it was likely due to my GPU drivers. However I did some additional tests afterward (cf first note), which seem to show there is a problem on modern systems also, even if it doesn't produce clear memory increase but just big page faults increase (hence performance decrease I guess?). I would have changed it to normal priority if I could... Anyway, my question wasn't meant to urge anyone, just to get some feedback as I'm not even sure it is considered as a defect. |
|
Kloot (developer) 2013-04-16 21:21 Last edited: 2013-04-16 21:25 |
It's unusual but not necessarily a defect, since you said that 1) the problem isn't triggered by all replays, even with bigger number of players and/or units 2) the big increase in page faults rate is observed from the start So let's say the problem is triggered only by maps with "many" features on them. Any game (or replay) on such a map should then consistently reproduce it and a simple modification of the healthbars widget (s.t. features are ignored) should be enough to prove or disprove this hypothesis. Here's what you do: the next game(s) you play on a suspect map, disable BA's unit_healthbars.lua and use the attached unit_healthbars_mantis3660.lua instead. Measure the memory usage and page fault count and compare both to the unmodified version. If there is a signicant difference, then we can decide to look further. |
|
bibim_ (reporter) 2013-04-16 22:16 |
1) & 2): maybe the problem is triggered not only by reclaimable features of the map, but also by units corpses. So replays containing lots of unreclaimed corpses would produce it more than replays containing lots of units but where all corpses are reclaimed fast. I don't have time right now to play a game and do once again all the measurements (I didn't get why I should wait to play a game to do the tests, instead of testing with a known replay producing the problem btw). However I quickly tested this new HealthBars widget with the replay I used for previous tests ( http://planetspads.free.fr/spads/tmp/20130328_214620_Tangerine_94.sdf ), and the difference is seen from the first seconds: this new widget behaves correctly regarding page faults. |
|
Kloot (developer) 2013-04-16 22:44 |
First assumption would be easy to test, load a featureless map like Comet Catcher and issue /cheat /give 500 armflash_dead. Replays will also work of course. If you could now remove lines 822 ("--[[") and 842 ("--]]") and report back how it behaves then, that would be enough information to determine if the issue is caused by the rendering or by the update portion of the widget. |
|
bibim_ (reporter) 2013-04-16 22:56 |
When removing these lines, the widget produces lots of page faults from the start. |
|
Kloot (developer) 2013-04-16 23:13 Last edited: 2013-04-16 23:15 |
Then most likely the GetVisibleFeatures function is to blame, but that did not change much between 91 and 94. Are you able to compile Spring yourself? |
|
bibim_ (reporter) 2013-04-16 23:57 |
No, I don't have any operational environment for the moment to build current Spring version for Windows. |
|
jK (developer) 2013-04-17 19:02 |
please test with 2nd done commit things to check for: * just run Spring 1h and see if LuaUI is still active after that |
|
bibim_ (reporter) 2013-04-20 14:14 |
"First assumption would be easy to test, load a featureless map like Comet Catcher and issue /cheat /give 500 armflash_dead" I just did the test and it triggers the high page faults rate problem too. |
|
Kloot (developer) 2013-04-23 20:33 |
Issue has been addressed for 95. Reopen if any recent development build (of the past 5 days) still shows these symptoms. |
| Notes |
| Issue History |
|||
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2013-03-29 14:40 | bibim_ | New Issue | |
| 2013-03-29 19:22 | bibim_ | Note Added: 0010315 | |
| 2013-03-29 19:30 | bibim_ | Note Edited: 0010315 | View Revisions |
| 2013-03-29 19:36 | bibim_ | Note Edited: 0010315 | View Revisions |
| 2013-03-29 19:55 |
|
Note Added: 0010316 | |
| 2013-03-31 01:03 | jK | Changeset attached | => spring develop cc709bf5 |
| 2013-03-31 01:03 | jK | Note Added: 0010332 | |
| 2013-03-31 04:07 | abma | Description Updated | View Revisions |
| 2013-03-31 04:08 | abma | Assigned To | => jK |
| 2013-03-31 04:08 | abma | Status | new => feedback |
| 2013-04-03 20:30 | abma | Note Added: 0010379 | |
| 2013-04-03 20:30 | abma | Note Edited: 0010379 | View Revisions |
| 2013-04-04 15:43 | bibim_ | Note Added: 0010386 | |
| 2013-04-04 15:43 | bibim_ | Status | feedback => assigned |
| 2013-04-16 16:32 | bibim_ | Note Added: 0010481 | |
| 2013-04-16 19:20 | Kloot | Note Added: 0010482 | |
| 2013-04-16 20:13 | bibim_ | Note Added: 0010483 | |
| 2013-04-16 21:17 | Kloot | File Added: unit_healthbars_mantis3660.lua | |
| 2013-04-16 21:21 | Kloot | Note Added: 0010484 | |
| 2013-04-16 21:25 | Kloot | Note Edited: 0010484 | View Revisions |
| 2013-04-16 21:27 | Kloot | File Deleted: unit_healthbars_mantis3660.lua | |
| 2013-04-16 21:27 | Kloot | File Added: unit_healthbars_mantis3660.lua | |
| 2013-04-16 22:16 | bibim_ | Note Added: 0010485 | |
| 2013-04-16 22:44 | Kloot | Note Added: 0010487 | |
| 2013-04-16 22:56 | bibim_ | Note Added: 0010488 | |
| 2013-04-16 23:13 | Kloot | Note Added: 0010489 | |
| 2013-04-16 23:15 | Kloot | Note Edited: 0010489 | View Revisions |
| 2013-04-16 23:57 | bibim_ | Note Added: 0010492 | |
| 2013-04-17 19:00 | jK | Changeset attached | => spring develop 848099e3 |
| 2013-04-17 19:02 | jK | Note Added: 0010498 | |
| 2013-04-20 14:14 | bibim_ | Note Added: 0010511 | |
| 2013-04-23 20:33 | Kloot | Note Added: 0010549 | |
| 2013-04-23 20:33 | Kloot | Status | assigned => resolved |
| 2013-04-23 20:33 | Kloot | Resolution | open => fixed |
| 2013-04-23 20:33 | Kloot | Assigned To | jK => Kloot |
| Issue History |