1 / 29

Power-efficient server provisioning in server farms

Anshul Gandhi (Carnegie Mellon University) Varun Gupta (CMU), Mor Harchol-Balter (CMU) Michael Kozuch (Intel, Pittsburgh). Power-efficient server provisioning in server farms. Motivation. Server farms are important for today’s IT infrastructure (Amazon, Google, IBM, HP, …)

chessa
Télécharger la présentation

Power-efficient server provisioning in server farms

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. Anshul Gandhi (Carnegie Mellon University) Varun Gupta (CMU), Mor Harchol-Balter (CMU) Michael Kozuch (Intel, Pittsburgh) Power-efficient server provisioning in server farms

  2. Motivation • Server farms are important for today’s IT infrastructure (Amazon, Google, IBM, HP, …) • However, server farms cost a lot of money to power ($4 billion in 2006) Server Farm Requests

  3. High-level problem statement • How many servers, given request rate ? • Don’t want to waste power Server Farm Requests

  4. Outline • Server farm model • Provisioning for fixed arrival rate • Provisioning for unpredictable, time-varying arrival rate • Future work

  5. Server farms IDLE servers consume a lot of power ~ 60 % of BUSY

  6. Server farms Turn IDLE servers OFF to save power HOWEVER

  7. Setup cost To turn on an OFF server .. • BUSY • OFF • SETUP • Time delay (setup time) • 1 min – 5 mins • and • Power penalty • peak power during setup time

  8. Setup cost To turn on an OFF server .. • BUSY • OFF • SETUP Should we ever turn servers OFF ?

  9. Server model • Server states: BUSY PBUSY 240 W IDLE PIDLE 150 W OFF POFF 0 W SETUP PSETUP 240 W • Setup times: TOFF→ON 200 s TON→OFF 0 s ON • Intel Xeon E5320 • 2 X 1.86 GHz quad-core • 4GB memory

  10. Server farm model • Poisson arrival process: λ(t) requests/sec • Exponentially distributed job sizes: E[S] secs • Load: ρ(t) = λ(t) ∙ E[S] Minimum # servers to handle incoming load Server Farm Requests FCFS

  11. Metric • Interested in response time and power conumption • Perf/W = 1/(Mean RT X Mean Power) • Maximize Perf/W

  12. Outline • Server farm model • Provisioning for fixed arrival rate • Provisioning for unpredictable, time-varying arrival rate • Future work

  13. Provisioning for fixed arrival rate • Existing solutions: prediction based, reactive controllers. • Is there a simple, yet, near-optimal solution ? Poisson arrivals Server Farm Max. Perf/W

  14. NEVEROFF • Keep n servers always ON (M/M/n) • Servers are BUSY or IDLE

  15. Perf/W for NEVEROFF

  16. INSTANTOFF • Turn servers OFF when IDLE • Servers are BUSY, OFF or in SETUP Auto-scales if n is high

  17. Perf/W for INSTANTOFF

  18. NEVEROFF vs. INSTANTOFF TON→OFF < γ E[S]/√ρ

  19. Near-optimality • Best of {NEVEROFF, INSTANTOFF} is optimal for single-server • Multi-server ? For ρ > 10, we are within 20% of OPT

  20. Outline • Server farm model • Provisioning for fixed arrival rate • Provisioning for unpredictable, time-varying arrival rate • Future work

  21. Unpredictable, time-varying demand • Data center demand has daily variations • INSTANTOFF can auto-scale

  22. Unpredictable, time-varying demand • NEVEROFF requires continual updates based on predicted load • Predictions are not always accurate • Can we find a simple traffic-oblivious policy? • Auto-scaling in nature

  23. DELAYEDOFF • Like INSTANTOFF, except we wait for twait seconds before turning IDLE servers OFF • Routing ? MRB routing is crucial !

  24. twait • Rule of thumb:twait ∙ PIDLE = TOFF→ON ∙ PON

  25. Near-optimality Worse at higher frequencies

  26. Auto-scaling capabilities • 1998 World Cup Soccer trace (ITA)

  27. Outline • Server farm model • Provisioning for fixed arrival rate • Provisioning for unpredictable, time-varying arrival rate • Future work

  28. Future work • Experimental evaluation of proposed schemes • Preliminary experiments on 15-server testbed using CPU-bound workload and sinusoidal arrival pattern • Experimental results agree with analysis • Web workloads: • What does the experimental setup look like ? • Try out various arrival traces and workloads

  29. Thank You! • Anshul Gandhi, Varun Gupta, Mor Harchol-Balter, Michael Kozuch Optimality analysis of energy-performance trade-off for server farm management, PERFORMANCE 2010 • Anshul Gandhi, Mor Harchol-Balter, Ivo Adan Server farms with setup costs, PERFORMANCE 2010 • Anshul Gandhi, Varun Gupta, Mor Harchol-Balter, Michael Kozuch Energy-efficient dynamic capacity provisioning in server farms, CMU technical report CMU-CS-10-108

More Related