多維性能數(shù)據(jù)預(yù)測及異常的研究.pdf

1、Resource availability often changes continuously in the cloud environment. There-fore，schedulers demand resource prediction help to make productive scheduling deci-sion. Comparatively，the scheduler needs a long time load

2、 prediction information to guide scheduling process because in cloud computing tasks usually take a significant amount of time to execute，even if most proposed prediction approaches are directed to one step ahead or shor

3、t term load prediction. As sometimes resources may be distributed and heterogeneous，those proposed methods could not lead to the good outcome when making long-term prediction. Most of the resource prediction methods were

4、 designed with the goal of getting the prediction much accurate. However，in some cases，priorities are given to long-term prediction.Anomaly detection techniques have been developed for some specific fields，while others w

5、ere developed for general purpose. By applying those techniques we can judge and differentiate whether a data is representing a normal or anomalous behavior. This happens when an anomaly detection approach define a regio

6、n that is demonstrating a normal behavior and provides a region where the data does not belong to the defined normal area as anomaly. Some techniques detect anomalies in an unlabeled test data set under supposition that

7、the majority of instances in the data are normal by investigating instances. Other techniques demand a data set that has been labeled as normal or abnormal and use training as classifier. There are also other techniques

8、that construct a model showing normal behavior from given normal training data set.Researches have been carried out in different features (CPU load，Memory us-age，disk spaces and network tra?c). Many prediction models wer

9、e evolved to predict single or multidimensional data. In this research，we will mainly focus on finding out new models and analyze prediction behavior from different multi-resource and deter-mine the highest accuracy. In