; Ganesh
Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

Ganesh

VIEWS: 1 PAGES: 6

  • pg 1
									Ganesh Ananthanarayanan
      Mentor: Randy Katz
               CS 294-14
Motivation/Objective
 Power consumption of datacenters is
  disproportionately high compared to the number of
  machines it contributes to the Internet
 “Green Grid” formed in April 2006
   “A group of technology industry leaders form The Green
    Grid to help reduce growing power and cooling demands
    in enterprise datacenters”
 Reduce power consumption of the networking
 component of datacenters
Related Work
 Wired Networks
    Wake on LAN/WAN
    Ethernet Adaptive Link Rate – Christensen and
     Nordman
 Wireless Networks
    Power Saving Mode (Wi-Fi) – Access Point buffers the
     packets
    Wake on Wireless – Multi radio Wireless LAN
 Sensor Networks
    Nodes elect a “leader” among them and proxy for each
     other
Proposed Approach
 Traffic prediction module forecasts “low-activity” periods
    How dynamic should this be?
 Suite of power saving techniques
    PSM – switch can buffer packets for a machine
    Pick low-speed mode for switch (10Gbps vs. 100 Mbps)
    Shut down individual ports, line card
    Collaborative power-saving
    Alternate “light-weight” network
        Control and low-bandwidth data
        Wireless? Low-powered Ethernet?
 Decision Framework – Power Model, Latency, Bandwidth
 Evaluation of power savings vs. performance
Risks/Data requirement
 Realistic workload traces
    Suggestions? 
 Power consumption models
    What is the power saved by de-energizing a port?
    What is the power saved by de-energizing a line card?
 Latency numbers
    How fast can I wake up a link?
Plan (fluid)
 Week 8: Get the latency and power values
 Week 9: Build the traffic prediction module
 Week 10: Design the various power-save techniques
 Week 11: Come up with a generic decision framework
  Week 12: Collect network utilization logs – synthetic or
  real workloads
 Week 13: Evaluate/plot the performance vs. power-save
  trade-off, Project Report

								
To top