应用于并行绘制系统负载均衡的改进随机森林算法

A Load Balancing Algorithm for Parallel Rendering System Based on Improved Random Forest

在并行绘制系统中,影响其性能的一个关键要素在于各个绘制节点是否负载均衡.基于屏幕空间划分的并行系统中,采用随机森林预测各个子屏的负载,通过预测结果调整屏幕的划分方式,来达到负载均衡的状态.而随机森林的预测准确度,决定系统达到负载均衡状态的时间.改进之后的随机森林算法,通过为决策树的叶子节点设置权重,并在预测过程中实时更新叶子节点权重,使得预测性能好的叶子节点对最终的预测结果产生更大的影响,从而提高随机森林的预测性能.实验证明,改进随机森林相比于原始RF,具有更高的预测准确性.

In a parallel rendering system,a key factor affecting its performance is whether each rendering node is load balanced.In a parallel system based on screen space division,a random forest is used to predict the load of each sub-screen,and the division of the screen is adjusted by the prediction result,so that the parallel rendering system reaches a load balanced state.The accuracy of random forest prediction deter mines the time it takes for the system to reach a load-balanced state.The improved random forest algorithm sets the weight of leaf nodes in the random forest and updates the leaf node weights in the prediction process in real time,so that the leaf nodes with good prediction perfor mance have a greater impact on the final prediction result.Experiments show that the improved random forest has higher prediction accura cy than the original RF.