Dynamic Virtual Resource Management In Clouds Coping With Traffic Burst

As a brand-new computing model, cloud computing has lead to a great revolution in the industry. With advantages like high scalability, professional management and so on, data centers help enterprises reduce a lot of costs on cluster maintaining and energy charge, as well as meet the needs of enterprises’ fast growing business development. However, for the cloud service providers, how to organize and manage the huge amount of both software and hardware resources in the data center is becoming a more and more importance problem. On the other hand, with the development of social media, information propagates with a faster speed. As a result, some web applications will attract a highly increasing number of visitors in a short interval of time due to propagation of some hot social events. So it becomes a vital problem that how to gurantee these web applications’ availability when they are suffering traffic burst. For the first problem, there are already some researches focusing on it, but if we put these two problems together, we want to figure out some solutions that could deal with traffic burst situations by utilize the scalability of data center.In this paper, we will start from the perspective of cloud service providers, and put forward a model, which manages virtual resource in clouds dynamically coping with traffic burst situations. On the one hand, this model will do resource management based on prediction, which solves the delay problems of traditional methods. On the other hand, this model will concertrate on keeping the users’ applications with a high availability when their visitors increase sharply in a short time. What’s more, we also do some work on performance tuning and cost saving.The work of this paper could be divided into several parts. Firstly, to deal with the traffic burst pattern, we put forward a hybrid prediction model based on Gompertz curve in our work. We use a predefined threshold to divide the flow curve into two parts: flat stage and burst stage. We use different prediction methods for each stage to get the number of visitors in the next time interval. Secondly, a resource management model is provided in our work. It is consisted of three sub models: the vm provisioning model, the vm placement model and the resource collection model. We use a queue theory related algorithm in the vm provisioning medel, which convert the number of visitors into the amount of resource required to meet service quality. In the vm placement model, which could be regarded as a variant of multi-objective programming problem, we list availability, performance and cost as the most important three factors and propose a greedy algorithm to solve this problem. Our model could achieve a relative high availability as well as a good performance and a low cost. In the resource collect model, we put forward a low-priority, lazy resource collect algorithm, which will help to reduce the active server number while maintain the service level.Simulation-based experiments are designed and implemented in this paper to test the effectiveness of our solution. As the experiment result shows, the Gompertz curve based combination prediction model generates a good result when applied to real data. Meanwhile, the resource management model proposed in this paper can achieve a high availability as well as reduce the cost of spending, which will be valuable to cloud service providers.The main contribution of this paper is that this is the first time research combine the common traffic burst situation with the resource management in cloud computing together. The scalability of data center is utilized to tackle the problem of traffic burst, which helps service providers to achieve high availability, and prediction method is used to overcome the problem of operation latency.