Hi Robert, my serialized sizes are: Ascii format: 137 byte binary format uncompressed 217 byte (!) binary format compressed 107 byte.
6 Month ago, I implemented 2 prototypes (TCP, UDP) . My result was, that TCP is not as fast as UDP, but it was the only possibilty to (nearly) sync the different channel's swap at final draw callback. On UDP I currently use broadcast, but thats not mandatory. Because of its unreliability, my UDP approach is fire and forget. So the slave tries to get as much as possible packets (to avoid framentation, the datacontainer must be smaller than the max payload of an UDP datagram) and uses the newest one for processing. My TCP approach was to send the frame information at the beginning to all slaves, then waits for all slave to report as ready, and if every client is ready, the sends the swap command to all clients. This solution was quite slow, so at the moment I focus an a UDP solution. I don't know which functionalities you would include in a cluster Node. Only transporting a transport container over network, or also complete frame sync? In my application I use to callbacks, eventdraw callback and finaldraw callback. At these two positions in the frame, the callbacks act depending of the role as master, slave or as standalone. They send information to slaves, receive information from master or do nothing. To make the cluster transparent to the scenegraph independent from the role, I implemented a dataIO interface which they query. Beyond this interface, there is the cluster driver (slave) or cluster driver + externalLink driver (master). The ladder connects the visual to its external control source, which can be a flighsimulator, a control program or anything else. My current transport content is: - viewmatrix - variables which are used of my objects updater classes - list of "executer" command, which can transport a command as enum and some parameter (e.g. to change weather at all clusternodes) This is my current approach. My project will be openscource in some days/weeks (depends on my time to set up my trac environment). Because you are under heavy load, maybe it is best If I continue my project, and when its published, you can take a look on it and we can discuss how man of the architecture could be abstracted, refactored to your wishes and included as cluster framework... Thank you! Cheers, Torben ------------------ Read this topic online here: http://forum.openscenegraph.org/viewtopic.php?p=24437#24437 _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org