a
Corresponding author: yym_sy@163.com
Research on the Prediction Model of CPU Utilization Based on ARIMA-BP
Neural Network
Jina Wang
1
, Yongming Yan
2,a
and Jun Guo
2
1
Liaoning Software Testing Center, Shenyang, China
2
College of Information Science & Technology Engineering, Northeastern University, Shenyang, China
Abstract. The dynamic deployment technology of the virtual machine is one of the current cloud computing research
focuses. The traditional methods mainly work after the degradation of the service performance that usually lag. To
solve the problem a new prediction model based on the CPU utilization is constructed in this paper. A reference
offered by the new prediction model of the CPU utilization is provided to the VM dynamic deployment process which
will speed to finish the deployment process before the degradation of the service performance. By this method it not
only ensure the quality of services but also improve the server performance and resource utilization. The new
prediction method of the CPU utilization based on the ARIMA-BP neural network mainly include four parts:
preprocess the collected data, build the predictive model of ARIMA-BP neural network, modify the nonlinear
residuals of the time series by the BP prediction algorithm and obtain the prediction results by analyzing the above
data comprehensively.
1 Introduction
In the recent cloud computing research field, the virtual
resource management problem has become a hotspot. The
VM dynamic deployment technique is one of the key
virtual resource management technique. Many experts
have done a lot of research about it.
At present, the study of the VM resource dynamic
deployment focus on the fine-grained resource
adjustment strategy. The current workload data and the
future change trend data are needed if a predictive and
accurate resource adjustment plan is to be made. Daniel
A Menasce
[1]
put forward a dynamic resource adjustment
plan following the changing workload by the CPU
priority or the CPU shares. ZHAO Weiming
[2]
presented a
new theory to improve memory utilization by predicting
memory usage of each VMs. Anton Beloglazov
[3]
proposed a set of the heuristic algorithms to select the
objective VM for migration when the VM violate the
SLA. A new distributed management strategy select the
VM whose CPU usage rate is the maximum in the
abnormal server in the reference 4. If after migrating the
CPU usage rate still be high, it would continue to migrate
the VM whose CPU usage rate is the maximum in the
rest VMs until recovering the server. Compared with the
reference 4, some researchers put forward a new
algorithm to select the VM for migration by predicting
the workload trend so that it can avoid the instantaneous
peak load
[5]
.
2 The Research Method of the CPU
Utilization Prediction Process
With the ongoing changes of the workload, the service
performance should be ensured by adjusting the resource
of the VM. After adding the predictive method, the VM
dynamic deployment plan can avoid the performance
fluctuation caused by the lag of the optimization process.
By the historical data a predictive model of the future
CPU utilization can be set up. With the predictive
information, more supports can be received by the
dynamic deployment plan.
So far, the method of the weighted average to
predicted single resource change trend is widely used in
the existing models, but the method using the composite
model to predict the CPU utilization by the time series
analysis technique is adopted in few researches. The
research emphasis in the former is to design the
weighting efficient system. A good weighting efficient
system means an accurate predicted result. But with
much subjectivity and arbitrariness it is very difficult to
design a good system. By researching, it is found that a
time series process consists of a linear structure and a
nonlinear structure that means the change trend rule can
be found easily by the time series.
The linear prediction model is mainly adopted in the
time series prediction method so that the method cannot
process the nonlinear data well. Compared with the
traditional data processing algorithms, BP neural network
is an effective nonlinear modeling method. It has great
DOI: 10.1051/
03009 (2016)
I
,
matecconf/2016MATEC Web of Conferences
65
650
CNSCM 2016
3009
© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the
Creative
Commons
Attribution
License 4.0 (http://creativecommons.org/licenses/by/4.0/).