Routing and scheduling for cloud service data centers

Cloud service data center;load balancing;queueing analysis;task assignment;data locality

Xie, Qiaomin

关键词: Cloud service data center;    load balancing;    queueing analysis;    task assignment;    data locality;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/42168/Qiaomin_Xie.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

In recent years, an increasing variety of dynamic-content web services, such as search, social networking and on-line commerce, have been moved to the Cloud service data centers. One of the features offered by Cloud data centers is on-demand service. To achieve this, current Cloud data centers are designed with an excess of provision for highly dynamic work load. Problems with such design include low server throughput and lack of scalability, which are considered as very important challenges for attaining system efficiency. This research aims to develop novel algorithms for the Cloud data center to achieve good performance while maintaining cost and energy efficiency. In general, Cloud service data centers consist of the front-end and back-end systems. To ensure a good level of service performance, neither the front-end nor back-end system should be neglected at the design of a Cloud data center. This study investigates features and challenges for Cloud service data centers. For the front-end system, the distributed design of load balancers is highly favored for achieving scalability. A novel algorithm is proposed for large-scale load balancing with distributed dispatchers. Both analysis and simulation show the advantage of the proposed algorithm over the state of the art.In the back-end system, intensive data processing is required to search, analyze and mine the vast amount of data. Cluster computing systems, like MapReduce and Hadoop, have provided an efficient platform for large scale computation. This research studies the data locality problem for cluster computing systems, which significantly affects system throughput and job completion time. In particular, a new task assignment algorithm is proposed and shown to significantly outperform the current implementation.