70 likes | 277 Views
Power aware scheduling/workload placement over cloud. CSE 591/494: Topics in Green Computing and Communication (Spring 2011 ) Zahra Abbasi. Topics. Introduction on cloud and virtualization Power aware workload placement within a virtualized data center
E N D
Power aware scheduling/workload placement over cloud CSE 591/494: Topics in Green Computing and Communication (Spring 2011) Zahra Abbasi
Topics • Introduction on cloud and virtualization • Power aware workload placement within a virtualized data center • Power aware workload across data centers
Power aware workload placement within a virtualized data center (1/2) • Server consolidation[Drexel-Springer08] • Objective: Minimizing number of physical machines • Decisions: How many active physical machine is required?, How (many) VMs should be assigned to a physical machine? • Resource allocation for VMs collocated in a single physical machine or for applications spanned across multiple VMs [HP-Lab,Eurocompsys09 ] • Objective: Minimizing performance degradation and (power consumption) of applications • Decisions: What is CPU share, memory share of VMs collocated in a single physical machine?
Physical organization: Each node hosts multiple applications running on VMs. Applications can span multiple nodes [HP-Lab,Eurocompsys09 ] Logical controller organization: Each application has one application controller. Each node has one node controller that arbitrates the requests from multiple application controllers
Power aware workload placement within a virtualized data center (2/2) • Power Capping [VirtualPower, VGreen,Coordinated power and performance control] • Objective: Achieving a power capping goal in a server cluster hosting VMs • Decisions: How VMs can be allocated to minimize power consumption? How power mode of servers should be controlled to achieve the power capping goal] • Taxonomy:
Power aware workload placement across data center(1/2) • Heterogeneity of data centers • Temporal and spatial variation of electricity price • Different power performance of data centers • Energy proportionality of physical servers • Data centers’ cooling system • Heterogeneity of network (bandwidth cost) • Workload placement schemes across data centers • Workload distribution across data centers (for stateless applications) • Application hosting management across data centers (for statefull applications)
Power aware workload placement across data center (2/2) • Taxonomy