| With the continuous upgrading of computer hardware equipment and the maturity of cloud computing technologies,cloud computing technologies and applications have achieved vigorous development in recent years.The cloud services are convenience and high efficiency.They have maked many research institutions and enterprises to invest in cloud computing.Cloud computing service provider sell computing resources.Meanwhile they also neet to decide how to define service charges.Rationality pricing method can greatly affect the user's choice.Amazon EC2 and Google GCE define detailed pricing methods.Cloud providers are continuously enriching the billing methods of cloud platforms.Such as Spot instances of Amazon is a dynamic pricing instance.Dynamic pricing instances is an kind of Cloud instance.The prices of These instances change in real time,such as Amazon's auction instance named Spot.The prices of auction instances arc usually much lower than on-demand and reserved instances.Another difference between dynamic pricing instances and on-demand instances is that the pricing of on-demand instances is stable and unchanged.While using dynamic pricing instances to obtain price concessions,the changed prices of dynamic pricing instances also make it more difficult for users to use cloud computing services.Users need to bid for the auction instance based on experience.A reasonable optimized bid price requires users to estimate the instance price in the future.However,it is difficult to predict the price in the future for many cloud users to optimize their own costs,and it is still confused about how to use dynamic pricing cloud instances to get the maximum price discount.It's still confusing to use this kind of dynamic pricing for cloud computing instances and when to apply for a Spot instance to get the maximum price discount.In this paper,we first propose a prediction model based on GRU network to predict the price of the dynamic instance spot instance of the Amazon Cloud using historical price data.Then we use differential evolution algorithm to optimize the bidding strategy of the instance based on the dynamic pricing of the IaaS.The main work is summarized as follows1.First,this paper analyze the price distribution of dynamic pricing instance.By using a self-defined parameter k-AMSE to quantify the spot instance price fluctuations,k-AMSE is better than MSE(mean square Error)int reflecting recent data fluctuations.We also statistically analyze the relevant influencing factors of price fluctuations.Then,we propose a price prediction model based on GRU network to predict the spot instance price.We evaluated the price prediction algorithm and compared its accuracy with other methods by RMSE(root mean square error)measurement.The experimental results show that our model has achieved a high accuracy rate for price prediction2.Strategic optimization of dynamic pricing instance combination and bidding.First,we explained the characteristics of dynamic pricing instance and the dynamic pricing auction model.We would like to mininize the cost of dynamic instance bidding and modeled the problem of dynamic pricing instance bidding and multi-instance combination as a constrained optimization problem.Then,we introduce the principle and detailed steps of the commonly used maximum bidding method and differential evolution algorithm to solve the optimization problem.Then,we improved the basic difference algorithm,proposed a differential evolution algorithm based on probability and adaptive parameters,and used this algorithm to optimize our combination of price bidding and instance use.Finally,we use the dynamic pricing instance price data published by the real Amazon platform,and use the maximum bidding,differential evolution algorithm,and adaptive differential evolution algorithm to optimize the data for bidding.Experiments show that our adaptive differential evolution algorithm obtains a better bidding strategy than the other two methods. |
PDF Full Text Request