Virtual network embedding for hybrid cloud rendering in optical and data center networks

作     者：Weigang HOU Lei GUO 

作者机构：School of Computer Science and Engineering Northeastern University State Key Laboratory of Networking and Switching Technology State Key Laboratory of Information Photonics and Optical Communications 

出 版 物：《Science China(Information Sciences)》 (中国科学:信息科学(英文版))

年 卷 期：2016年第59卷第2期

页      面：139-152页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 081201[工学-计算机系统结构] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported in part by Open Foundation of State Key Laboratory of Networking and Switching Technology(Grant No.SKLNST-2013-1-08) Open Foundation of State Key Laboratory of Information Photonics and Optical Communications(Grant Nos.IPOC 2013B001 IPOC2014B009) Fundamental Research Funds for the Central Universities(Grant Nos.N140405005 N130817002 N130404002) Foundation of the Education Department of Liaoning Province(Grant No.L2014089) National Natural Science Foundation of China(Grant Nos.61401082 61471109 61302070) Liaoning Bai Qian Wan Talents Program and National High-Level Personnel Special Support Program for Youth Top-Notch Talent 

主　　题：hybrid cloud rendering optical and data center network virtual network embedding mapping cost lower bound 

摘      要：Animation rendering consumes massive computation time, therefore cloud rendering is emerging as a solution. Cloud rendering runs over the Data Center Network(DCN) and consolidates heterogeneous DC resources into a single cloud renderfarm, where plentiful computing resources can sufficiently accelerate any rendering process. And if one user wants to get a quick animation result, a high-speed optical interconnection is an urgent requirement, thus cloud rendering needs a convergence of Optical and DCN(ODCN) as the substrate network. In the ODCN supporting cloud rendering, each rendering task will be successfully handled only when we embed its virtual network into the cloud renderfarm. But because a virtual network includes virtual machines and virtual lightpaths, we must simultaneously perform the node-level mapping between virtual machine and server, as well as link-level mapping between virtual lightpath and fiber link(s). In addition, the joint implementation of the Photorealistic cloud Rendering(PR) and Non-Photorealistic cloud Rendering(NPR) should be considered to exhibit the unique animation effect with the low mapping cost. In this paper, considering the unique characteristic of hybrid cloud rendering, we flexibly select routing strategies according to the rendering task type. We then utilize server consolidation and traffic grooming to achieve node- and link-level mappings,respectively, thus building a mapping-cost-aware cloud renderfarm that includes multiple virtual networks. The mathematical formulation is also made with a bound analysis. Especially for the lower bound, we analyze the least number of servers and wavelengths(i.e., mapping cost) consumed by hybrid cloud rendering. In terms of heuristics, according to the processing order of rendering tasks, Smaller Virtual Resource First(SVRF) and Manycast Routing First(MRF) algorithms are proposed by us. In SVRF, NPR tasks are first tackled and then PR tasks follow. MRF is a reverse process of SVRF. The simu