More and more clouds with diversified properties have been built. Many of them span multiple geographical locations over the globe, imposing time-varying costs on and offering different service proximities to users. Clouds could be private or public, requiring different levels of administration effort and providing different levels of freedom to control. Hybrid clouds, which blend together multiple public and private clouds, possess properties of both types. Based on these diversified properties, multiple clouds have the potential to provide services with higher scalability, lower operational cost and better QoS. To exploit this potential, I examine the means of deploying services on multiple clouds, that can maximize utility in dynamic environments.
Firstly, I consider the migration of an important representative application, content distribution services, to geo-distributed hybrid clouds. I model the problem of joint content data migration and request dispatching as a unified optimization framework, and then design a dynamic control algorithm to solve it. The algorithm bounds the response times within the preset QoS target, and guarantees that the overall cost is within a small constant gap from the optimum that can be achieved by a T-slot look ahead mechanism with known future information.
Secondly, I study the problem of efficient scheduling for disparate MapReduce workloads on hybrid clouds. I build a fine-grained and tractable model to characterize the scheduling of heterogeneous MapReduce workloads. An online algorithm is proposed for joint task admission control for the private cloud, task outsourcing to the public cloud, and VM allocation to execute the admitted tasks on the private cloud, such that the time-averaged task outsourcing cost is minimized over the long run. The online algorithm features preemptive scheduling of the tasks, where a task executed partially on the on-premise infrastructure can be paused and scheduled to run later. It also achieves such desirable properties as meeting a pre-set task admission ratio and bounding the worst-case task completion time.
Thirdly, I consider a cloud computing resource market where a broker is employed that pools the spare resources of multiple private clouds and leases them to serve external users' jobs. I model the interaction between the broker and the private clouds as a two-stage Stackelberg game. As the leader in the game, the broker decides on the pricing for renting VMs from each private cloud. As a follower, each private cloud responds with the number of VMs that it is willing to lease. Combining all this with Lyapunov optimization theory, I design online algorithms for the broker to set the prices and schedule the jobs on the private clouds, and for each private cloud to decide the numbers of VMs to lease. The broker achieves a time-averaged profit that is close to the offline optimum with complete information on future job arrivals and resource availability, while each private cloud earns the best that it can.
Through theoretical analysis and empirical study, I rigorously examine the cost or profit optimality, and QoS guarantee of my design, and show that they can indeed outperform existing solutions. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy
Identifer | oai:union.ndltd.org:HKU/oai:hub.hku.hk:10722/206669 |
Date | January 2014 |
Creators | Qiu, Xuanjia, 邱炫佳 |
Contributors | Lau, FCM, Wu, C |
Publisher | The University of Hong Kong (Pokfulam, Hong Kong) |
Source Sets | Hong Kong University Theses |
Language | English |
Detected Language | English |
Type | PG_Thesis |
Rights | The author retains all proprietary rights, (such as patent rights) and the right to use in future works., Creative Commons: Attribution 3.0 Hong Kong License |
Relation | HKU Theses Online (HKUTO) |
Page generated in 0.0105 seconds