1 / 28

Design and Analysis of an Energy Agile Cluster Computing System

Design and Analysis of an Energy Agile Cluster Computing System. Andrew Krioukov , Prashanth Mohan, Stephen Dawson-Haggerty, Sara Alspaugh , David Culler, Randy Katz. Grid Evolution. renewable, variable, intermittent, greatly non-dispatchable. non-renewable, reactive, dispatchable.

azana
Download Presentation

Design and Analysis of an Energy Agile Cluster Computing System

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Design and Analysisof an Energy AgileCluster Computing System Andrew Krioukov, PrashanthMohan, Stephen Dawson-Haggerty, Sara Alspaugh,David Culler, Randy Katz

  2. Grid Evolution renewable, variable, intermittent, greatly non-dispatchable non-renewable, reactive, dispatchable mostly dispatchable Supplies Ideal Future Old Grid Today Loads oblivious, stochastic, mostly non-power proportional reactive, mostly power proportional oblivious, flat

  3. Grid Evolution renewable, variable, intermittent, greatly non-dispatchable non-renewable, reactive, dispatchable mostly dispatchable Supplies Ideal Future Old Grid Today Loads oblivious, stochastic, mostly non-power proportional reactive, mostly power proportional oblivious, flat

  4. Grid Evolution renewable, variable, intermittent, greatly non-dispatchable non-renewable, reactive, dispatchable mostly dispatchable Supplies Ideal Future Old Grid Today Loads oblivious, stochastic, mostly non-power proportional power proportional, reactive, grid-aware oblivious, flat

  5. Pieces Needed SUPPLIES: provide power communicate renewable availability, price Internet LOADS: adapt demand communicate forecast Grid electricity information ?

  6. Renewable Integration Non-dispatchable, variable supply Figure of merit: amount of wind used. How do we get here? Power proportional, grid-aware loads Pacheco wind farm Scientific computing cluster NREL Western Wind and Solar Integration Study Datasethttp://wind.nrel.gov/Web_nrel/

  7. dispatchable supply Power oblivious, flat load Time power proportionality grid-awareness

  8. Data Center Loads 5,000 servers at Google average 30% utilization data center consumption dominated by IT load IT load driven by workload need power proportionality need load shaping mechanism IT equipment is not power proportional power (W) utilization Pelley, et. al, Understanding and Abstracting Total Data Center Power, 2009 Barrosoet. al. The Case for Energy-Proportional Computing, 2007 SPECpower Results http://www.spec.org/power_ssj2008/results/power_ssj2008.html

  9. Power Proportionality Spinning Reserve

  10. Architecture

  11. Outline • Motivation • Enabling technology • Methodology • Algorithms • Evaluation

  12. Renewable Energy Component

  13. Formulation Option 1: grid blend (open system) Option 2: dedicated wind farm (closed system) Other Wind Requires assuming load is negligible fraction of grid – not realistic Fit load to specific wind farm We assume the wind farm is sized for the data center. http://www.greenhousedata.com/

  14. Wind Wind power over 48 hours from a wind farm in Monterrey County, California. Variation in wind power for month long intervals at multiple wind farms.

  15. Workload Component

  16. Workloads Interactive: Latency sensitive, generally short jobs e.g., web app server, email server, etc. Wikipedia traffic Request Rate Batch: Less latency sensitive, longer jobs e.g.,analytics, scientific computing Torque jobs Num Jobs

  17. Slack slack = max run time – job duration

  18. Slack in Real Systems Cluster: NERSC Franklin Average duration: 98 min Average slack: 68 min Cluster: EECS PSI Average duration: 55 min Average slack: 17 hours

  19. Grid-Aware Batch Scheduling • example goal: shape load to match wind availability • method: exploit temporal slack Pacheco wind farm Scientific computing cluster

  20. Greedy Algorithm B(t) = power budget for next 10 min Sort jobs by slack Schedule all jobs with no remaining slack Schedule other eligible jobs in least-remaining-slack order until B(t) is exceeded

  21. Grid-aware scheduling increases wind energy use. Run-immediately, grid-oblivious scheduler Greedy, grid-aware scheduler Correspondingly, reduces grid dependence.

  22. Reduction in grid dependence is robust to choice of wind farm.

  23. As slack increases, grid dependence diminishes. Franklin PSI

  24. Grid-aware scheduling is equivalent to 5 hours worth of data center-sized batteries.

  25. Grid-aware scheduling is equivalent to 5 hours worth of data center-sized batteries. 4

  26. Summary • Power proportionality and grid-aware scheduling • Energy savings, renewable integration, grid stability reduce grid dependence by half equivalent to 5 hours of batteries • Next steps slack in other systems ...?

  27. The End Questions?

More Related