They are chosen based on their proximity and the load of the node. Each node of the system hosts and answers queries on a group of objects in a zone, which is the minimal axis-aligned rectangle. We use a distributed spatial index on top of a self-adaptive tree structure. Kwery performs efficient reverse geolocation queries on large numbers of moving objects updating their position at arbitrary high frequencies.
The optimal number of avatars per CPU and the performances of our system have been evaluated simulating tens of thousands of avatars connecting to a Kiwano instance running across several data centers, with results well beyond the current state-of-the-art.We also propose Kwery, a distributed spatial index capable to scale dynamic objects of virtual worlds. The load is constantly balanced among Kiwano's nodes which adapt and take in charge sets of avatars according to their geographic proximity. The avatar-to-avatar interactions and related computations are then bounded, allowing the system to scale. In Kiwano we employ the Delaunay triangulation to provide each avatar with a constant number of neighbors independently of their density or distribution. However, surprisingly, virtual worlds are still unable to host simultaneously more than a few hundred users in the same contiguous space.The main contribution of the thesis is Kiwano, a distributed system enabling an unlimited number of avatars to simultaneously evolve and interact in a contiguous virtual space. Virtual worlds attract millions of users and these popular applications -supported by gigantic data centers with myriads of processors- are routinely accessed.
0 kommentar(er)
