Power Considerations in Virtualized Environments Swetha P.T. - PowerPoint PPT Presentation

Power Considerations in Virtualized Environments Swetha P.T. Srinivasan supervised by Prof. Umesh Bellur Indian Institute of Technology, Bombay Research and Innovation Symposium for Computer scientists (RISC 2016) April 2 nd 2016

Background 1. Power consumption of a server P total = P static + P dynamic (Watts) 2. Energy consumption E = Σ t i =1 P ( i ) (Watt-hour) 3. Dynamic Voltage and Frequency Scaling (DVFS) - combination of voltage and f set for processor

Goal: Energy Optimal VM Provisioning Goal: Energy-optimally provisioning VMs on DVFS-enabled servers Ensure completion time SLAs are met

Where Existing Heuristics Fail 6 VMs need to be placed Best Fit Decreasing / First Fit Decreasing: ordering by CPU requirement size VM1 VM3 VM5 VM2 VM4 VM6 Idle VM# CPU% Time(s) 60 100 vm 1 (a) PM1 PM2 100 100 Power (W) 60 100 vm 2 75 75 50 50 200 30 vm 3 25 25 0 0 200 30 vm 4 0 100 200 300 400 0 100 200 300 400 Time (s) Time (s) 400 10 vm 5 400 vm 6 10 (b) PM1 PM2 PM3 100 100 100 Power (W) 75 75 75 85% energy; 50 50 50 More saving if 25 25 25 0 0 0 DVFS applied 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 Time (s) Time (s) Time (s) Hypothesis - tightest packing (least #servers) is not energy efficient for deadline-based provisioning.