面向能耗优化的云数据中心资源分配问题研究-templecis.pdfVIP

Let’s Stay Together: Towards Traffic Aware Virtual Machine Placement in Data Centers Xin Li, Jie Wu, Shaojie Tang, Sanglu Lu Nanjing University Temple University 2014.05.01 Outline  Background and Scenario  Problem Statement  Homogeneous Case  Heterogeneous Case  Conclusion 2 Background  Virtual machine (VM) placement  Tenants submit their resource requirements to the cloud system, and the cloud decides how to implement the resource allocation.  One of the primary task in virtualization- based cloud system.  The cost is one of the major concerns for the cloud providers.  PM-cost  N-cost Scenario  We use slot to represent one basic resource unit. (CPU/memory/disk)  Tenants submit their resource requirements, in terms of the number of VMs (slots).  Each slot host one VM  For one tenant, it could be one project group, and each VM can be assigned to one group member.  The VMs (group members) finish the task cooperatively. Virtual Machine Placement  Inter-PM traffic  Inter-VM traffic  The objective  Minimize the total inter-PM traffic. How to determine the communication cost? 5 Communication Model  Two communication models Centralized Distributed Model Model 6 Communication Cost  The traffics between VMs are assumed to be aware in most related works.  Here, we do NOT adopt this assumption.  We focus on network cost, measured by the number of traffic links between VMs  One request may be placed on m