Resource Allocation Methods For Load-Balance Over Cloud Computing Environment

In recent years,with the development of the Internet and information technology,cloud computing has received extensive attention from the industry and academia.The expansion of the scale of cloud platforms,on the one hand,brings users more cloud computing services with greater computing capacity.On the other hand,it exacerbates the increasingly prominent energy consumption problem of cloud platforms.Many studies have shown that the energy consumption of cloud platforms is closely related to the utilization of their resources.Low resource utilization will cause energy waste,which not only increases the operating costs of the platforms but also violates the concept of green cloud computing.At the same time,by improving load-balance of cloud platforms,it will also help improve stability and resource utilization,and reduce energy consumption.Recently,more and more high performance computing(HPC)applications have abandoned traditional operating methods and selected lower-cost pay-as-you-go cloud services.In this scenario,the resource management of cloud platforms faces many challenges:1)For the static scenario that resource requirement information of all HPC applications before running is known,how to use the known information to calculate resource allocation scheme in advance to improve resource utilization and load-balance for cloud platforms.2)For the dynamic scenario where users may submit HPC application at any time,how to allocate resources for the newly submitted HPC applications based on the current idle resource status to ensure high resource utilization and load-balance of cloud platforms.Based on the aforementioned challenges in resource allocation for cloud platforms running large-scale HPC applications,this paper focuses on this scenario and conducts related research on load balancing methods for cloud computing resources.Specifically,the main work of this paper includes the following aspects:1)To illustrate the relationship between HPC applications and cloud platforms running HPC applications,this paper first proposes a layered system framework that combines them.Considering that HPC applications can be subdivided into one or more service requirements,which the framework defines as meta services,cloud platforms use cloud services to allocate resources for meta services and execute them.In order to carry out further design and analysis of the method,this paper builds the system model based on the framework,deduces the calculation formula of resource utilization and load-balance,which are performance indicators in this paper,and gives the target optimization function.2)In the static scenario,all the information of meta services which make up HPC applications is known before running the HPC applications.Based on this scenario,this paper proposes a method of static allocation for cloud computing resources on load-balance.The method is divided into four steps:meta service preprocessing,resource usage monitoring for cloud services,static resource allocation for meta service subset,and global static resource allocation method.The resources allocation schemata are calculated using the known resource requirements and time requirement information of all meta services.Then,the HPC applications are executed according to these calculated resource allocation schemata.This method allocates resources for meta services by selecting the cloud service with best load-balance,and optimizes load-balance according to the end time of meta services.This method can improve the resource utilization and load-balance of cloud platforms running large-scale HPC applications in static scenarios.3)In the dynamic scenario,there may be new HPC applications that need to be executed at any time.The running process of HPC applications can be seen as cloud platforms dynamically allocate resources for the newly arrived meta services.Based on this scenario,the paper proposes a method of dynamic allocation for cloud computing resources on load-balance.The method is divided into four steps:maintenance of assignable resource tables,dynamic resource allocation for the arrival of meta services,dynamic resource allocation for the end of meta services,and global dynamic resource allocation method.This method allocates resources for meta services by selecting the best cloud service on load-balance,and achieves re-balance when load-balance is reduced.It can effectively improve system load-balance while satisfying good resource utilization.