基于AR模型的主机负载预测

线性时间序列作为一种概率统计方法，已经被运用到各个领域中。AR模型是最常见的一种时间序列模型，是根据时间序列的自相似性质，利用时间序列在过去时刻的观察值推算时间序列的出现规律，预测今后可能出现的观察值。文中利用AR模型预测在分布式实时环境上主机的负载情况。对于有多台主机的分布式实时环境，当有一个新的任务到达时，如果能够较准确地预测出各台主机在今后的一段时间内的负载情况，调度器可以有选择地将任务分配给适当主机，不仅可以满足尽可能多数量的实时任务的最后时限，并且可以提高系统的性能。

Host Load Prediction Based on AR Module

As a probability and statistics methodology, linear time series has been applied to many fields. AR module is one of the most familiar linear time series, which based on selfsimilarity characteristic of time series, predict host load using past observed host load. This paper will use this statistical character and predicable characteristic of AR module to predict host load in distributed real - time system. In a system with several hosts involved, if the application can predict the exact load of each host, it will be helpful to not only the schedule but also the performance, for the scheduler can choose a host to perform.