Random freeze...?

[EJGr.Ost_Chamel]

We have made some quite different experiences. We have run 14 missions in the Guadalcanal sector ( http://by.sturmovik.de/seow/ejgr/Statis ... uadalcanal - we used the Guadalcanal template from the SEOW homepage) and I remember a mission, where there were about 95 planes in the air but everything ran without problems for more than an hour.

On the other hand a buddy stopped working on a template for the "Ostfriesland" sector, because several test missions all froze after about 15 minutes. Does anybody know, whether the Ostfriesland-Map is one of those, that are known to have a memory leak bug?

Greetings

Chamel

[BS8th_Bulau]

I have just run a test with this coop mission, on the Ostfriesland map. This coop has 76 aircraft, and probably at least that many ground objects moving. I entered >GC into chat at 2 minute and 1 minute intervals. Memory total reported as follows:

0:00 74mb
0:02 75mb
0:04 81mb
0:06 84mb
0:08 87mb
0:09 89mb
0.10 89.3mb
0:11 89.8mb
0:12 91.2mb
0:13 94.5mb
0:14 94.1mb

A few seconds later, the game locked up.

[IV/JG7_4Shades]

I have played about 5 missions of Tuchel Heath (with the map images changed to look like Poland) with Warg and some guys from II/JG77 and IV/JG7. We typically had about 6 human pilots, about 20-30 aircraft and 20-30 platoons moving. Game performance was a bit chunky, especially in the early missions as the front lines engaged over the rivers and bridges. We certainly had no lockups or freezes.

Cheers,
4Shades

[BS8th_Bulau]

Okay, that is a considerably smaller scale than we have been attempting. In fact, we have even found 40 aircraft (20 per side) to be somewhat limiting for the commanders, and adding the ground movements we are crashing this map. I have no idea if my test indicates a memory leak problem, or just too much objects for this particular map.

We are going to try another map, but we are not sure which at this point. We would like one that will support fairly large missions, like 50-60 aircraft flying, plus same number of ground movements. If this is excessive for any map, please advise.

Is there a correlation between the physical size of a map, and the number of objects/flights/movements it can support? Hmm, apparently not, since you list Ardennes as a map that you have run large campaigns on, yet it is one of the smallest maps.

[IV/JG7_4Shades]

Hi Bulau,

We are all really facing an IL-2 limitation here. SEOW, at its heart, simply creates FMB-style coops. So any SEOW campaign is limited in playability by the capacity of IL-2 to run the coops generated.

Setting up a decent SEOW campaign takes a bit of time, so to avoid fruitless effort it is often worthwhile doing some play-testing of large coops first, so you can find a workable combination of map on one hand, and numbers of flights/movements/stationaries on the other.

Smolensk is known to be a smooth map, as is Prokhorovka.

Please keep us updated with your findings.

Cheers,
4Shades

[rnzoli]

I entered >GC into chat at 2 minute and 1 minute intervals. Memory total reported as follows:
...
0:13 94.5mb
0:14 94.1mb

A few seconds later, the game locked up.

A tell-tale sign of the stupid memory bug in IL-2. The limit is at 96 MB (even if your PC has 4 GB). Over 90MB, you may already experience strange lagging, pulsed warping, as the underlying JVM is desperately trying to free any garbage memory, anticipating a total congestion coming on.

Normally maps use 25-50, or with heavy object population 60-70MB. But larger and more detailed maps definitely use more memory, event with same number of objects, than small ones.

There is a temporary workaround for 4.08, but I reported this bug to the dev team, so may I kindly ask you to test the same with 4.09 beta patch? The result would be interesting for many of us.

[BS8th_Bulau]

Yes, that describes, exactly, the behaviour in the last few minutes. For the record, I have tabulated the contents of the coop mission:

76 Aircraft
82 [Chiefs]
339 [NStationary]
140 [Buildings]
10 [StaticCamera]
121 [FrontMarker]

So that I'm comparing apples to apples, I have decided to copy and paste this mission onto the Prokhorovka and the Smolensk maps, and test run them in a similar fashion. Also, these maps come recommended as memory-friendly. I may also test some others, time permitting.

Zoltan, thank you for answering that question, and for your feedback on this. Yes, I can certainly test this also using the 4.09b patch.

[BS8th_Bulau]

I've had a chance to do a little testing of this. I've copied our big mission onto the Prokhorovka, Smolensk, and Moscow1 maps. The only editing I did was to make sure all the aircraft takeoff and landing waypoints were at an airbase.

Prokhorovka easily ran for an hour, starting out at 41.5 megs, and never rising higher than 47.2 megs total memory. At one hour, this mission is essentially winding down, and I saw no reason to continue any longer test.

Smolensk also ran for an hour without even working up a sweat, starting at 39.4 megs, and reaching a high of 40.8 megs, finishing the hour at 39.0 megs.

Moscow1 has not been run for the hour duration at this time. I started it, and checked memory use immediately, finding it to be using 40.2 megs. I expect that it would play out similarly to the foregoing two maps, and last the hour without incident.

As per the request from Zoli, I tested the original mission on the Ostfriesland map, using the latest 4.09 beta patch. In two attempts, it seemed to fare even worse than 4.08, starting at 73 megs, and lasting only 10 minutes before locking up, using a little over 88 megs at that point.

[IV/JG7_4Shades]

It would be interesting to have a list of base memory usage of each SEOW map, calculated with the tiniest template, e.g. a host seat and no other objects.

[BS8th_Bulau]

Yes, I had the same thought, 4Shades! As I was running these tests on loaded maps, I wondered what the blank Ostfriesland map would use, so I made one with a single A-20 aircraft. Now, I don't have my notes at hand at the moment, but it was, I recall, about 34 megs, (40 megs less than our big coop starts out at).

Of course, now I am also wondering what a blank Smolensk map would use. Would it be 40 megs less as well...probably not, as that would be practically zero! It would be very simple to check all the maps in the same fashion, at least to compare them. Such numbers would be of more value, if we also had a better idea of how memory use relates to numbers of objects and also how it changes as the mission progresses.

[BS8th_Bulau]

The results of my baseline memory checks of all SEOW maps, containing only a single P-11c aircraft. Map size is in grid squares, but I did not check if all map grids are same size in kilometres. Question marks denote where there is more than one choice of map, and I wasn't sure which one the SEOW MP uses.

Code: Select all

SEOW Name           IL-2 Name              Size         Memory (mb)
 
Ardennes            ArdennesSummer         F-4          20.3
Berlin              Berlin                 T-8          27.9
Crimea              Crimea                 BC-17        24.1
Cyrenaica           Desert Online          H-6          18.2
Guadalcanal         Guadalcanal Early      BX-46        17.9
Gulf of Finland     Gulf of Finland1?      BQ-21        42.9
Hawaii              Hawaii                 DV-155       22.7
Imphal              Burma                  N-10         21.2
Italy               Italy Online           P-16         26.3
Iwo Jima            Iwo Jima               DA-80        18.2
Khalkhin Kol        KalkhinGol             D-4          20.4
Kuban               Kuban                  V-20         23.6
Kurland             KurlandAutumn?         X-32         31.0
Kursk               Kursk                  S-17         27.5
Kyushu              Kyushu (South Japan)   BC-33        26.3
Lvov                Lvov                   S-17         26.6
Malta               Online5Summer          J-9          19.7
Marianas            Marianas               DV-90        19.8
Midway              Midway                 BX-58        18.9
Moscow              Moscow1                S-17         23.2
Murmansk            Murmansk               BE-26        24.4
New Guinea          New Guinea             BQ-24        21.7
Normandy            Normandy2              O-10         28.1
North Sea           Norway                 N-14         20.4
Okinawa             Okinawa                BD-30        20.4
Oslo                Online3Summer          E-5          20.6
Ostfriesland        Northwest Europe       J-8          35.3
Prokhorovka         Prokhorovka            K-10         23.7
Singapore           Singapore 2            BN-27        24.2
Smolensk            Smolensk               S-17         26.3
Stalingrad          Stalingrad1            S-17         26.3
Tubruq              MTO (4.09 map)         T-21         22.9
Tunis               OnlineMT               E-5          21.4

[IV/JG7_4Shades]

Interesting that Guadalcanal, known to be a problem map, has a pretty low initial memory demand.

Nice data.

4Shades

[II/JG3K.Brandle]

S!

Another test for you to check out:

Load game
START FMB
OPEN Console over water so you can read it

Look for ####House without collision### errors upon map startup.

You need to reload the game each time you test a map.

http://forums.ubi.com/eve/forums/a/tpc/ ... 6531077073

I dont know what this means but Smolensk, Moscow,Prok,Stalingrad,Lvov are all error free and they also provide the smoothest gameplay. Coincidence? I dont think so but you can see the attention span of the people in the ubi.zoo. FritzGriffin is a mouth breather and his responses show a Pavlovian approach to exchanging information on forums.

[BS8th_Bulau]

I recall seeing that sort of message when I started running a dedicated server, v2.04, a couple years ago. It always came up when loading a mission that used the Online4Summer map. I assumed it was an object I had placed on the map, but since it ran fine, I never investigated it. In fact, it's not really clear that it is an error message or not.

I was a little surprised to see those messages from other maps. For example:

Guadalcanal has three:
##### House without collision (3do/Buildings/Port/BaseSegment/live.sim
##### House without collision (3do/Buildings/Port/Floor/live.sim
##### House without collision (3do/Tree/Tree2.sim

Gulf of Finland has two:
##### House without collision (3do/Buildings/Port/BaseSmallSegment/live.sim
##### House without collision (3do/Buildings/Port/Floor/live.sim

Singapore has one:
##### House without collision (3do/Tree/Tree2.sim

Additionally, Singapore began generating another error message, starting once in the cockpit, and repeating continually at 3 sec. intervals:
WARNING: Bad Water if MapRID == 0

Our big coop generates other error message, which appear to be related to a particular aircraft used, such as:

INTERNAL ERROR: HierMesh: Can't find chunk 'Flap03_DO'
INTERNAL ERROR: HierMesh: Can't find chunk 'GearR9_DO'

Messages like that are generated periodically over the course of the hour long mission. I don't know which aircraft, flyable or AI, but it could be identified by process of elimination.

For my tests above, I did not restart the game for each map. I've noticed that, the numbers can vary by 0.2 to 0.3 mb for the same map, and it doesn't seem to matter if it is first map loaded, or subsequent.

Something not clear to me is what all this means, or rather, what it actually is telling us, if anything. The exact same mission appears to "bulk up" different maps by different amounts. For example, our big mission seems to "add" 40 mb to the Ostfriesland map, but only around 15 mb to the Smolensk and Prokhorovka maps.

I also did a simple calculation, not shown above, dividing memory use by map size in grid squares. I called this number "density" for lack of a better term, and I thought it might yield some useful data, but I'm really not sure. Some maps are HUGE, but consist largely of water. Guadalcanal, supposedly a problem map, has a density of 0.01 mb/grid, which is very low, while Prokhorovka, a "good" map, has a density of 0.22 mb/grid, a midrange value. The highest value for this is Khalkhin Kol at 1.28 mb/grid, then:
Ardennes 0.85 mb/grid
Olso 0.82 mb/grid
Ostfriesland 0.44 mb/grid
Cyrenaica 0.38 mb/grid

[II/JG3K.Brandle]

I think they are tied to the map. House and collision were words I focused on. After testing Wake or Midway I thought I might be able to track down on the map, where houses might be colliding and sure enough, 2 houses or something were overlapping. I think the game log keeps track of house destruction and my hunch is this is the problem. The more cities, the more collisions, the better the chances to have a problem.

I dont know anything about computer language and dont care about the technical aspects other than how to build a smooth mission. I think a breakdown of memory of maps coupled with the techincal aspects of SEOW targets like bridges, factories,airfields, fuel should be a determining factor in how we build maps.

Last but not least, you can preview your frame rates on a map by building your template, starting up the ingame fps meter and comparing the frames in game to in FMB. I have found the two to be tied together to a degree that makes it useful.


S!