Neural Network-Based Load Prediction for Highly Dynamic Distributed Online Games
Euro-Par '08 Proceedings of the 14th international Euro-Par conference on Parallel Processing
Enhancing Grids for Massively Multiplayer Online Computer Games
Euro-Par '08 Proceedings of the 14th international Euro-Par conference on Parallel Processing
| Hi-index | 0.00 |
A novel neural-based solution to the problem of dynamicload balancing in homogeneous distributed systems isproposed. The winner-Take-All (WTA) neural networkmodel is used for implementing the selection andlocation policies of a typical dynamic load balancingalgorithm. Unlike most of the previous literature thatassumed independent tasks, which is not always true,tasks with interprocess communication requirements areconsidered. All delays due to any usage of thecommunications network resource are taken intoaccount. A simulation study was carried out to verify theeffectiveness of the proposed approach, results werecompared against the no load balancing case. Althoughperformance improvements are dependent on the systemoverall load, load intensity per node, and nature of tasks,the results suggest that it is always beneficial to use loadbalancing than not at all.