An Approach for Cloud Resource Risk Prediction

3 downloads 305 Views 219KB Size Report
that didn`t occur in the enterprise private cloud, such as the resource default risk ... Micro economics principal-based cloud bank model is a kind of public cloud .... value and quality of resource difference value in the t-th observation period, ...
Available online at www.sciencedirect.com Available online at www.sciencedirect.com

Procedia Engineering

ProcediaProcedia Engineering 00 (2011) Engineering 29 000–000 (2012) 3292 – 3296 www.elsevier.com/locate/procedia

2012 International Workshop on Information and Electronics Engineering (IWIEE)

An Approach for Cloud Resource Risk Prediction a*

a

b

Hao Li , Miao Xin

b

Dept. of Software, Yunnan University, Kunming, China Dept. of Information Science and Engineering, Yunnan University, Kunming, China

Abstract The resource risk management is a key issue in cloud computing, which is directly related to the reliability and security of the cloud computing platform. In the cloud bank model [1], almost all resources come from messes of cloud resource providers, who are distributed on the Internet. This feature caused some special resource risk types that didn`t occur in the enterprise private cloud, such as the resource default risk caused by people or machines. Analyzed the features of cloud computing and the cloud bank model, this paper delivers a cloud resource risk prediction model, which is based on mathematical statistics. This method advances the risk management strategy from the accident response to the beforehand prediction.

© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Harbin University of Science and Technology Keywords:cloud computing; cloud bank model; risk prediction; mathematical statistics;

1.

Introduction

Cloud computing is a style of commercialized distributed computing in which dynamic scalable virtualization resources are provided as services to users over the Internet [2]. Resources are distributed in physical, but are servicing in form of a single entity finally. Most cloud computing infrastructure architecture are such business models as IaaS (Infrastructure as a Service), PaaS (Platform as a Service) and SaaS(software as a Service). The commercialization of cloud computing makes the distributed computing technology and economics principals joined together. Prof. Buyya has proposed an economic-based grid computing model [3], and posed some original opinions in cloud resource scheduling and management. The cloud bank

*

* Corresponding author. E-mail address: [email protected].

1877-7058 © 2011 Published by Elsevier Ltd. doi:10.1016/j.proeng.2012.01.482

Hao Li and Miao / Procedia Engineering (2012) 3292 – 3296 Author nameXin / Procedia Engineering 00 29 (2011) 000–000

2

model in this paper is also on the basis of economics principal, and have referenced the real commercial banks` risk management strategy in the aspect of cloud resources management. The cloud resource management and risk control is the key technology in cloud computing. Different from the grid computing, the cloud computing platform is operating on the unreliable nodes, and the nodes default are regarded as normality. Current technology is implemented by node redundancy and node automatic passing on, which is the default response technology after the default has happened. But it dependents on the high-speed Intranet in the enterprise data center. However, as the application background of this paper, the cloud bank model is an open public cloud computing model. The resources and ownership are physically distributed on the Internet. Comparatively, the network is unreliability and the network cost is higher. So, although the cloud bank model has also adopted the redundancy and node passing on strategy, but it can not be completely depended. A better strategy is to make a prediction beforehand. 2. Cloud Bank Model Micro economics principal-based cloud bank model is a kind of public cloud computing model. In the aspects of resource organization, it adopt the “deposit-loan” pattern, whose architecture is quite similar to the real commercial bank. There are three basic roles in the cloud bank model: the Cloud Resource Provider (CRP) , the Cloud Bank (CB) and the Cloud Resource Consumer (CRC). The CRPs provide computing resources to a CB, and get profit from it. The CB provide kinds of IaaS services to CRCs. In the whole system, almost all resources the computing environment needed come from messes of CRPs. By means of the resource visualization technology (implemented by Cloud Resource Agency (CRA))[4], the CRP resources will be joined into a unified visualization resource pool who is managed by the CB. Depending on these resources, the CB make the scheduling according to an economics principal [5] and provide kinds of IaaS services for CRC. Being a core component of cloud bank model, the resource risk management will affect the stable operation and the reputation of one cloud bank. As a more open IaaS model, the resources in cloud bank model are more distributed in physical. For the resource do not physically belong to one cloud computing facilitator, the cloud bank have to face an accident which may occur at any time: the resource default during the contract (first line in table 1), caused by varieties of reasons. This situation is an important type of resource risk in cloud bank model. Owing to this resource accident usually happen during the contact period, this type of risk is defined to be the resource default risk. The resource default risk is the key point this paper focused on. In the cloud bank`s system, there runs a Cloud Resource Risk Monitor (CRRM) whose function is to real-time monitor the resource quantity and quality in the visualization resource pool and send the risk coping steatery (shown in table 2) order when it is needed. To prevent the resource default risk as far as possible, the cloud resource risk monitor need a suit of approaches to predict the risk beforehand according to some variables with significant meaning. What we need is to finally build a cloud bank resource risk measuring model for the CRRM. In the following parts of the paper, we will definitely analyze the reason causing resource default. Then, according to these reasons, we will transform them into the risk measuring variables for the risk prediction model. And finally, base on a mathematical statistics method, the default risk model will be built. This paper also contain the simulation experiment and the summary. 3.

The Reason Caused Resource Default Risk

Macroscopicly speaking, there are two kinds of resource default in cloud bank model: the first type is defined to be the subjective reason-caused default, and the second type is defined to be the objective reason-caused default.

3293

3294Subjective Hao Li and Miao Xin / Procedia Engineering 29 (2012) 3292 – 3296 3.1 Factors-caused Default

Subjective factors-caused default is such a kind of default that the default is artificially induced by the people who hold the resource entity. The subjective factors-caused default during the contract, or artificial-reasons default, means that the people (resource providers) sign out from the CRA on their own initiative (e.g., the resource provider shut down the agent). The CRP must sign a contract with CB before the providers` resource join into the resource pool, and the contract contains the penalty provision. Consequently, considering the penalty cost and the credit rating`s being reduced, the cost of default is quite expensive for CRPs. So, the most probably default motivation is the goal of chasing a higher earnings, after the appearance of a higher resource purchase price by another cloud bank in the market. Based on the analyzing above, the default profit value can well measure the subjective factors-caused default reason. 3.2 Objective Factors-caused Default Different from objective factors-caused default, objective factors-caused default is commonly induced by machine fault or Internet failure. Objective factors-caused default is mostly caused by some random accidents such as internet failure, power failure, or computer virus outbreak in different areas. These accidents will induce a large number of resources loss in a short time and even causes service quality shock. Some certain indicators such as network delay, resource working historical credit record and response delay, can well reflect the resource real-time situation and also with statistical significance. The cloud bank compile these variable into the a standardized quality of resource value [6] in form of Service-Level Agreement(SLA)[7] according to their practical needs, and this value can measure objective factors-caused default possibility. 4.

The Resource Risk Prediction Method in Cloud Bank Model

After the analysis above, this paper delivers a cloud resource default ratio (DR) prediction method on basis of the multiple linear regression analysis[9]. As the reasons leading to resource default can be classified into objective factors-caused default and subjective factors-caused default, the prediction model contains at least two explanatory variables (binary regression). The explanatory variables are formalized as follows: • Default Profit (M): Default profit value is formalized subjective factors-caused default variable. Intuitively, its meaning is the benefit the CRP can get if making a default. The computational formula is as follows: M = (Imax - I) - B

(1)

I is the resource purchase price bided by the cloud bank the CRP in current contract, and Imax is the

highest bid among all cloud banks in the market. So, (Imax - I) is the resource price discrepancy. B is the default penalty specified in the contact. • Quality of Resource Difference Value (R): The computational formula is as follows: m

~

R= Q−

1 ∑ Qj m j =1

(2)

~

Qj is the resource quality evaluation value, which is obtained by benchmark testing. Q is the resource quality reference value set by cloud bank, which is set to be the ideal resource quality. The value`s setting reflect the resource quality requirement of one cloud bank. For the reason that it must prevent the large amount of resource loss in a shout time, a variable is needed to represent the average level of the resource

3295

Hao Li and Miao / Procedia Engineering (2012) 3292 – 3296 Author nameXin / Procedia Engineering 00 29 (2011) 000–000

4

1 m ∑ Qj quality in the whole visualization resource pool. m j =1 is the average resource quality value, which can well reflect the general state. Both of the two variables are positively correlated with the default ratio. After obtained explanatory variables, we can get the formalized possibility of default model based on binary linear regression. The regression equation as follows:

Pˆt = b0 + b1C t + b2Rt

(3)

Pˆt

C

R

is DR`s (default rate) predictive value of the t-th observation period. t and t are default cost value and quality of resource difference value in the t-th observation period, which are the explanatory

b0 b1 , and b2 are regression parameters. b Regression parameters ( 0 , b1 and b2 ) can be solved by the following linear equations:

variables of regression equation.

n n ⎧n ⎪∑ Pt = nb0 + b1 ∑ M t + b2 ∑ Rt =t 1 =t 1 ⎪ t =1 n n n ⎪n 2 ⎨∑ M t Pt = b0 ∑ M t + b1 ∑ M t + b2 ∑ M t Rt = ⎪ t 1 =t 1 =t 1 =t 1 n n n ⎪n 2 ⎪∑ Rt Pt = b0 ∑ Rt + b1 ∑ M t Rt + b2 ∑ Rt = ⎩ t 1 =t 1 =t 1 =t 1

(4)

The cloud bank should accumulate the historical data in a certain length period, and make it into a

historical data table. When it need to make a prediction to the (t+1)-th period, every

M t and Rt in the b0

historic data table are substituted into the equations (5), then we can obtain the regression coefficients:

,

b1 and b2 . Then substituted the regression coefficients and (t+1)-th explanatory variables ( M t +1 and Rt +1 ) into the equations (4), we can get the predictive value at last. The whole progress of resource default risk prediction and response method is listed as follows: 1) Risk Prediction Stage:

M t and Rt ) are substituted into the formula (5). ′ ′ M t +1 R Then the regression parameters can be solved; then the current and t +1 are substituted into the ′ Pˆ Before the (t+1)-th period`s start, the historical data (

formula (4), and the DR prediction value ( 2) Risk Response Stage:

t +1

) can be generated.

′ ′ Pˆt +1 is the important parameter to active the risk response mechanism. The Pˆt +1 means the severity ′ M t +1 of resource default. If the is the significant influencing factor, the cloud bank should preferred ′ R modify softness response strategy(pricing strategy); else if the t +1 is the significant influencing factor, the cloud bank should active the hardness response strategy (start reserved resources).

3296 Experiment 5.

Hao Li and Miao Xin / Procedia Engineering 29 (2012) 3292 – 3296

With the help of the CloudSim, we have written a program to simulate the running of CRRM. This program has simulated a computing cluster with 50 virtual machines. And it also has been configured the bandwidth and network topology. The final result is: When the amount of history data we used is 100 periods (n=100), the average fitting degree is 70.56%. The maximum is 83.17%, and the minimum is 58.69%. However, after the further data arranging, it is found that it is not positive correlation between the amount of history data and the fitting degree. For example, we have separately fetched 10, 20, 30……100 periods data to make the same experiment. The best average fitting degree is arise when the amount of history data is 60. And the maximum average fitting degree is 75.83%. 6.

Summary and Future work

This paper analyzed the resource management in cloud computing environment, and delivered a economics principal-based cloud computing model. On basis of resource risk in cloud bank model and commercial bank risk management methods, this paper delivered a cloud resource management prediction method, which have solved a bottleneck problem in for the cloud bank model. However, there are still some issues to be solved, such as the cloud resource quantization, and the automatic mapping between risk intensity and response strategy. These issues are the emphasis in future research. Acknowledgments. The research reported herein was supported by the National Nature Science Foundation of China 2010 (No.61063044) and the Key laboratory in Software Engineering of Yunnan Province under Grant No.210KS05. References [1]

Hao Li, YZ, Joan Lu,Xuejie Zhang, and Shaowen Yao, “A Banking .Based Grid Recourse Allocation Scheduling”, GPC

Workshops, 2008. [2]

Wikipedia, Cloud Computing, http://de.wikipedia.org/wiki/Cloud_Computing

[3]

Rajkumar Buyya, Economic-based distributed resource management and scheduling for Grid computing, in Thesis. 2002,

Monash University. [4]

Adabala, S., V. Chadha, P. Chawla, R. Figueiredo, J. Fortes, I. Krsul, A. Matsunaga, M. Tsugawa, J. Zhang, M. Zhao, L.

Zhu, and X. Zhu, From Virtualized Resources to Virtual Computing Grids: The In-VIGO System. Future Generation Computer Systems, 2004. [5]

Hao Li Huixi Li, “A Research of Resource Scheduling Strategy For Cloud Computing Based on Pareto Optimality M×N

Production Model”. [6]

Hao Li,Guo Tang,Joan Lu ,Shaowen Yao, The QoS Resource Quantification Based on the Grid Banking Model,

Proceedings of the First International Conference on 'Networked Digital Technologies' (NDT 2009) Ostrava, The Czech Republic, July 29 - 31, 2009.ISBN:978-1-4244-4615-5.pp.548-553 (published by IEEE). [7]

H. Ludwig, A. Keller, A. Dan, R. King, R. Franck: A Service Level Agreement Language for Dynamic Electronic

Services. Electronic Commerce Research (3), Nr. 1, pp. 43 –59, Kluwer Academic Publishers, 2003. [8]

Kai Song , Hao Li, “The Research of cloud resource management Model Based on Economics”, IEEE, Conference on

Computer Design and Applications, 2011. [9]

Wikipedia, Regression analysis, http://en.wikipedia.org/wiki/Regression_analysi