Hi, Sérgio Miguel Elias Cabrita wrote: > Hi, i've been doing some tests to the the opensg 1.6, in a display wall > (4x3 configuration). I've found a couple of things: > > -The tutorial for the clusterServer doesn't free memory after the client > is finished. Because we're benchmarking the wall we run several tests > from a script and after a few run's the server's started to crash. We > suspected some sort of memory leak, and after some manual checking with > ps, i found that the server's process eats up about 2gb of memory after a > few run's displaying some hi-res models. We're a bit new to this (opensg) > but we tried to see if there was a way to plug the leak... we didn't find > many references, so we tried to destroy the glut window, and the > ClusterServer and create new ones when the cliente finishes. This almost > worked, but the memory alocated for the model still remained. Does anyone > know a good way to fix this memory leak? For now, we just finish the > server process when it detects that the client finishes and with a bash > script lauch it again in a endless loop. > > -When running on the wall, we made a test where the model started centered > inside a tile, and moved in a straight line threw all the other tiles. On > high resolution models, there is a small delay when the model hit's the > tile's boundry. This delay is considerable, specialy on models with > textures. We have a small model, about 3000 triangules, with textures that > hits the boundry, stop's for about 1 second and then moves on. Is this a > bug? feature? problem? Any known way to solve this? > > this delay is a result of frustum culling. Geometry end Textures are send to the graphics card when they are visible. To avoid this delay we are switching off frustum culling for the first frame. With this the whole geometry is send to the graphics card on all hosts, independent from visibility. Frustum culling can be enabled and disabled in the RenderAction.
> -We noticed that 1.6 also brought BalancedMultiWindow Class. Correct me if > i'm wrong, but this is supost to work like the MultiWindow class, except > that less loaded server's get the load of more loaded server's, maximazing > the speedup. Is this correct? If so, could some one give me an example of > how it work's, because we used it, but the model isn't displayed properly > like in the MultiWindow Class. We just replaced the declaration of the > type of window. Do we have to do something else? We would like to try this > new class and see the results. > > Yes. the BalancedMultiWindow should work as the MultiDisplayWindow. You can check balancing by setting setShowBalancing(true). It is important to set the following fields. size: Width and Height of the whole Display setHServers: number of horizontal displays setVServers: number of vertical displays. If you set setBalance(false) it should work in the same way as the MultiDisplayWindow. It is possible to balance the load between display servers and it is possible to add additional servers in the servers list. Marcus ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 _______________________________________________ Opensg-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/opensg-users
