Minimizing The Resource Reservation Cost For The Cloud Broker

In recent years,with the development of cloud computing technology,there are many IT companies began to provide IaaS services across the world.For cloud service providers,they always charge users in two ways: on-demand virtual machine instances and prepayment-reserved virtual machine instances.The on-demand virtual machine instance has higher price for the unit time and is suitable for users who need to use virtual machines for a short period of time.The reserved virtual machine instance has lower price for the unit time but has longer effective time,and is suitable for users who need to use virtual machine instances for long period of time.For the pricing mechanism of the cloud service providers,there are two main problems: for most common users,they can only select the on-demand virtual machine instance with higher unit price;for the people who buy the prepayment-reserved VMs,the VMs they bought are not always in the processing,which will result in waste of resources and cost increment.Based on this reality,the cloud broker,a middle agent between the cloud service providers and users,appeared.And its main function is to buy a large number of virtual machine instances from the cloud service providers,and then in the form of time-multiplexing,these virtual machine instances are subleased to the average user at a price lower than the cloud service providers.For cloud broker,how to reserve virtual machine instances from the cloud service providers to meet users' needs while minimizing his own costs of resource reservation is an urgent problem to be solved.In the scenario with one cloud service provider,an offline and an online algorithm are designed for the broker to schedule virtual machine instances.Based on the idea of time-multiplexing,the demand graph based on the users' tasks is constructed hierarchically.The main idea of the offline algorithm is based on the demand graph and the optional virtual machine instance with the lowest average effective price will be selected to cover the demand graph in the way of layer by layer.According to the Banhcard problem,the online algorithm uses the user's historical request data to determine the virtual machine instance to be bought at current time.By performing experiments on three different datasets,the broker can save 40% and 31% cost of the resource reservation based on the offline algorithm and the online algorithm respectively,which improves the profit space of the cloud broker.This paper designs an offline solution for the new application scenario that the cloud broker reserves VM instances from multiple cloud service providers in time delaying mode.The main idea of the algorithm is based on the skiing lease problem.First,the virtual machine instances with longer time and lower unit price are used to satisfy the user's requirements.Then,the virtual machine instance with shorter unit price and higher unit price is used to meet the user demand.For all of the alternatives,we select the policy with lowest cost.Compared with the offline algorithm for the cloud broker in the scenario with one cloud service provider,the broker can save 5% extra cost.