Reliability Analysis Of Cloud Platform Resources Status

As an emerging technology, cloud computing means that the computing capability is becoming a commodity to circulate through the network. Due to its low cost and openness, the cloud platform technology is widely used in business, military and academic fields. Thus the reliability of combining resources of cloud platform is particularly important. Effective cloud architecture can integrate computation and network resources to provide users with a reliable and high-availability environment, meanwhile it can also reduce unnecessary waste.The present representative technology to analyze the reliability of cloud platform resources is the cloud platform monitoring technology, which focuses on monitoring and displaying the running situation of the server and virtual machine of the cloud platform. It can provide the operator with great help for further maintenance.However, the current monitoring technology of cloud platform only monitor the usage of platform resources and trigger alert based on simple exception rules. The exception rule which is set by the experience and subjective judgments of managers is insufficiently credible. There is no systematic strategy to analyze cloud platform-oriented resource reliability in existing methods.In order to meet above requirements, this thesis proposes a method of reliability analysis of cloud platform resources status. In this design, time series model for each resources are created, then the clustering and correlation analysis technology, which are common algorithms of data mining and machine learning, are adopted to analyze all of above time series. The main work of this thesis including:1,dbscan clustering technology and k-means clustering technology are combined effectively,bring a new method called db-means.2, using DTW(Dynamic time warping), which is widely used in audio signal analysis, to adapt time series so that improve clustering result.This performs better than the Eculidean distance.3, Different time series clustering strategy are proposed to adapt to different working modes on server and virtual machine. Based on their historical working records, users can have a clear insight into the operation situation of entire system.4, The time sequence segment mode serves as dataset of correlation analysis to solve the continuous attribute discretization problem of cloud platform's time sequence data.5, Frequent item set mining is achieved through correlation analysis of time sequences, which contributes to discover the potential relationship behind various resources usage of cloud resources.The advantage of this design is that,it discover the normal condition and the hidden relationship between the resources of the cloud platform server and virtual machine. It provides a theoretical basis for setting task allocation strategy, finding system bottlenecks and setting warning rules,etc.