- 學位論文
在混合雲端環境下保證服務品質自動調整排程演算法
Adaptive Scheduling with QoS Satisfaction in Hybrid Cloud Environment
摘要
近年來雲端計算與服務大量快速的增加，使用者對於雲端服務的要求也越來越高，而目前現有的雲端技術與服務中並沒有對使用者保證QoS(服務質量)。現有雲端環境可分為兩大類：公有雲及私有雲。混合雲則是結合兩者的優點，基於安全性和建構完整性較高的私有雲，再利用需要收費的公有雲來處理短暫的突發流量和服務請求。如何在混合雲中提出一個基於服務質量保證、最大化私有雲使用率及最小化租用公有雲的花費則是我們主要的研究議題。 我們提出在混合雲端環境下保證服務質量自動調整排程演算法，利用任務執行時間預測以及動態規劃來達到最大化私有雲的使用率，並減少任務等待時間及任務執行時間，以增加私有雲處理更多任務的可能性；另外我們對於收費的公有雲制定了成本花費函數，藉由動態規劃計算及最小化成本花費函數來達到最低的成本花費；而使用混合雲的優勢也是減少不必要的硬體環境建置成本，我們將私有雲的使用率做最佳化的分配之後，也可以減少使用需收費公有雲資源的任務數量。
並列摘要
Cloud computing is an increasing research topic in the recent years. Cloud provides different types of services, such as PaaS, IaaS and SaaS. For an economical and efficiency way, hybrid cloud becomes an important environment. How to ensure QoS satisfaction in hybrid cloud is our main objective. We also need to maximum the utilization of private cloud and minimize the cost in public cloud. In this thesis, we have proposed an Adaptive Scheduling with QoS Satisfaction in Hybrid Cloud Environment. Combine both advantages of private cloud and public cloud: stable, security, flexibility, economically and pay-per-use, using hybrid cloud environment. It provides QoS demand for user, and guarantee job response time. Using runtime estimation and dynamic programming to achieve near-optimal allocation in private cloud and maximize the utilization and minimize the runtime of tasks. For critical inputs or overloading in private cloud, tasks should be selected to public resources. For the features in public cloud, pay-per-use, may only charge for the submit tasks in brief time. But we still try to minimize the cost of renting public slots. By better allocation on private cloud, scheduling can reduce the amount of tasks that need public slots resources. For the tasks have to be dispatched to public cloud, we choose minimal cost strategy based on the characteristic of tasks such as code size and information data size. As the experiment shows, our scheduling algorithm AsQ achieve better performance in reducing task waiting time, task runtime and task finish time than existing scheduling algorithm. In the same condition, AsQ can also guarantee more QoS satisfaction rate. Cost analysis shows for ensure deadline constraint, AsQ cost much less than others. In the experiment with COSHIC, AsQ presents better balancing between QoS and cost consideration.