Dynamic Capacity Planning For Virtual Servers

With the popularization and development of Internet technology in recent years,the amount of data and users on the network requests increased explosively,this growth also requires the server's processing power to make a corresponding increase.In today's typical data center,we need a computing services system with high reliability and high extendibility to improve network throughput.Given the current pace of development of hardware devices has lagged far behind the growth rate of users' needs,so how to utilize current services resource rationally has become a hot topic.Nowadays,a large number of companies begin using virtualization mechanisms to support the management of server clusters.The main motivation for using this technique is to increase the flexibility,and the server can be quickly reprogrammed to meet the needs of users better,and greatly reduce the overall cost.Therefore,we need to design the appropriate capacity planning models and tools to allocate server resources effectively.The main purpose of this paper is to propose a viable and efficient capacity planning model,which can plan for future server requirements,and provide dynamic allocation scheme.In this paper,the main research work and innovative results reflected in the following aspects:1.We explore how to generate reliable user requests to make capacity planning precisely.Since when we do capacity planning,we are not able to know the future load accurately and the change in the amount user requests.This problem requires us to design a program to produce the simulation data which are as close as possible with real data,and use our simulation data to implement capacity planning.First we adopt an effective prediction algorithm to predict future user requests.Then we use appropriate sampling method to extract the appropriate number of user requests from historical data as simulation user requests data.2.We discuss the capacity planning models and methods specifically.We proposed two capacity planning scenarios: static and dynamic capacity strategy.Static strategy fixed the mapping relation between servers and users,we need to design appropriate algorithms on historical data to make a fixed mapping which utilize server resource effectively,but will not cause the server overload frequently.The dynamic strategy does not require fixing mapping between users and servers.When a new user request comes in,we need to decide which server to render services based on the current server cluster load.Meanwhile,dynamic strategy will try to rebalance the loads among servers.3.We develop a system using above strategies.The user can use different strategies we mentioned to plan for the future the number of servers on the system.While the system also provides a large number of statistical analysis functions,they can be applied for the user to view the change trend on servers,user and overall performance in the historical data.