Cost Reduction and Quality-of-Service

The optimal workload distribution for a single day simulation is shown in Fig. 4.2(d) where the simulation data are as in Fig. 4.2(a)-(c). We can see that the optimal workload distribution is not uniform and some data centers receive more workload at different times of day. Let us examine the two scenarios marked in this figure.

For Scenario 1 between hours 7 and 8, the total incoming workload in the system is low, as we can see in Fig. 4.2(c). Therefore, it is possible to handle most of the workload using only wind power, making wind power availability the key factor in decision making. As a result, the majority of the workload is forwarded to Data Center 1 which has significantly higher wind power availability at this time of day, as we can see in Fig. 4.2(a). However, for Scenario 2 between hours 20 and 21, the total incoming workload in the system is high. Therefore, the wind power available is not enough to run data centers and we need to purchase electricity from the grid, making the price of electricity the key factor for decision making in this case. As a result, the majority of the workload is forwarded to Data Center 2 which has the lowest price at this time of day, as we can see in Fig. 4.2(b). Finally, we can see from the results in Fig. 4.3 that the optimal workload distribution scheme can stabilize queues and maintain short queue lengths in all data centers to assure quality-of-service.

Comparing to a uniform workload distribution plan that allocates workload among data centers equally, the optimal workload distribution scheme can significantly reduce the cost of running data centers as shown in Fig. 4.4(a). We can see that the proposed design outperforms uniform workload distribution at any time of day. Repeating the simulations for 10 different days, based on the World Cup 98 workload data from May 28, 1998 to June 6, 1998, we can see that the cost of running data centers will reduce by 29.2%, i.e., $8,797 on average, leading to a total saving of $87,970 over 10 days.

4.4 Conclusions

In this chapter, we proposed an optimization-based workload distribution framework for Internet and cloud computing data centers with behind-the-meter renewable

gen-0 1 2 3 4 5 6 7 8 9 1gen-0 11 12 13 14 15 16 17 18 19 2gen-0 21 22 23 24 50

100 150 200 250 300 350 400 450 500

Time of Day (Hour)

Queue Length (Number of Requests Waiting)

Data Center 1 Deta Center 2 Data Center 3 Queue Limit Parameter D = 500

Figure 4.3: The queue lengths are stabilized for the data in Fig. 4.2.

erators. In this regard, we took into account computer servers’ power consumption profiles, data centers’ power usage effectiveness, total workload in terms of the num-ber of service requests received at each time of day, instantaneous queue lengths at each data center, availability of renewable power at different locations, and price of electricity at different locations. Each data center is designed to adapt its power con-sumption to constantly follow the time-varying trend of renewable power generated while the number of service requests waiting at the data centers’ queues are moni-tored to assure quality-of-service. Using various experimental data, simulation results show that the proposed design can significantly reduce data centers’ cost of electricity while maintaining quality-of-service.

This chapter can be extended in several directions, For example, by considering incorporating service differentiation, statistical whether forecasting information, and propagation delays between user and data center locations into the problem formu-lation to achieve more accurate designs.

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0

0.2 0.4 0.6 0.8 1 1.2 1.4

Time of Day (Hour)

Instantaneous Total Cost (Dollars per Second) Optimum Workload Distribution

Uniform Workload Distribution Peak Cost Due to Lack of

Local Wind Power Generation

(a) Instantaneous Total Cost for a Sample Day

1 2 3 4 5 6 7 8 9 10

0 10 20 30 40 50 60 70 80

Day of Experiment

Total Daily Cost (Thousand Dollars)

Optimal Workload Distribution Uniform Workload Distribution

29.2% = $8,797 Average Cost (Uniform Workload Distribution)

Average Cost (Optimal Workload Distribution)

(b) Total Daily Cost Over 10 Days

Figure 4.4: Total cost of running all three data centers when the optimal workload distribution is compared with a uniform workload distribution: Instantaneous cost for the daily data in Fig. 4.2; (b) Total daily cost for simulating 10 days.

Chapter 5 Future Works

The results in this chapter can be extended in several directions. First, the cost model considered in this chapter can be extended to take into account cost elements other than energy cost. Second, given renewable energy forecasting models and day-ahead pricing tariffs, the proposed short-term energy and performance management can be further extended to daily or monthly planning, e.g., to decide whether increasing the number of available computer servers is economical. Finally, the obtained mathe-matical model in this chapter can be adjusted to also include potential profit if a data center participates in ancillary services in the power grid. This last aspect can particularly facilitate close interactions between data centers’ and utilities and grid operators in smart power grid systems.

Bibliography

[1] Z. Liu, M. Lin, A. Wierman, S. Low, and L. L. H. Andrew, “Geographical load balancing with renewables,” in Proc. of the ACM GreenMetrics Workshop, San Jose, CA, April 2011.

[2] L. Rao, X. Liu, L. Xie, and W. Liu, “Coordinated energy cost management of distributed internet data centers in smart grid,” IEEE Transactions on Smart Grid, vol. 3, no. 1, pp. 50 –58, March 2012.

[3] R. Katz, “Tech titans building boom,” IEEE Spectrum, vol. 46, no. 2, pp. 40–54, February 2009.

[4] H. Mohsenian-Rad and A. Leon-Garcia, “Coordination of cloud computing and smart power grids,” in Proc. of the IEEE Smart Grid Communications Confer-ence, Gaithersburg, MD, October 2010.

[5] M. K. S. Gurumurthi, A. Sivasubramaniam and H. Franke, “Reducing disk power consumption in servers with drpm,” Computer, vol. 36, no. 12, pp. 59–66, De-cember 2003.

[6] R. K. Megalingam, T. S. Hassan, P. Vivek, A. Mohan, and M. . Rao, “Power consumption reduction in cpu datapath using a novel clocking scheme,” in 2nd IEEE International Conference on Computer Science and Information Technol-ogy, August 2009.

[7] M. Kandemir, S. W. Son, and M. Karakoy, “Improving disk reuse for reducing power consumption,” in ACM/IEEE International Symposium on Low Power Electronics and Design, August 2007.

[8] R. K. Megalingam, B. V. Krishnan, M. Mithun, R. Srikumar, and V. V. Sarma,

“Gating and serializing the data path of cpu for low power consumption,” in International Conference on Parallel Processing Workshops, September 2009.

[9] J. Heo, D. Henriksson, X. Liu, and T. Abdelzaher, “Integrating adaptive compo-nents: An emerging challenge in performance-adaptive systems and a server farm case-study,” in Proc. of the IEEE International Real-Time Systems Symposium, Tucson, AZ, December 2007.

[10] E. Pinheiro, R. Bianchini, E. V. Carrera, and T. Heath, “Dynamic cluster re-configuration for power and performance,” in Compilers and Operating Systems for Low Power, M. K. L. Benini and J. Ramanujam, Eds. Kluwer Academic Publishers, September 2003.

[11] X. Wang and M. Chen, “Cluster-level feedback power control for performance optimization,” in Proc. of the IEEE Int. Symp. on High Performance Computer Architecture, Salt Lake City, UT, February 2008.

[12] Z. Karakehayov, “Dynamic clock scaling for energy-aware embedded systems,”

in 4th IEEE Workshop on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, September 2007.

[13] X. Fan, W. D. Weber, and L. A. Barroso, “Power provisioning for a warehouse-sized computer,” in Proc. of the ACM International Symposium on Computer Architecture, San Diego, CA, June 2007.

[14] J. S. Chase, D. Anderson, P. Thakar, A. Vahdat, and R. Doyle, “Managing energy and server resources in hosting centers,” in Proc. of ACM Symp. on Operating systems principles, Banff, Canada, October 2001.

[15] L. Rao, X. Liu, L. Xie, and W. Liu, “Minimizing electricity cost: Optimization of distributed internet data centers in a multi-electricity-market environment,”

in Proc. of IEEE INFOCOM, Orlando, FL, 2010.

[16] J. Li, Z. Li, K. Ren, and X. Liu, “Towards optimal electric demand management for internet data centers,” IEEE Transactions on Smart Grid, vol. 3, no. 1, pp.

183–192, March 2012.

[17] M. Ghamkhari and H. Mohsenian-Rad, “Optimal integration of renewable en-ergy resources in data centers with behind-the-meter renewable generators,”

in Proc. of the IEEE International Conference on Communications, Ottawa, Canada, June 2012.

[18] ——, “Energy and performance management of green data centers: A profit maximization approach,” Submitted to IEEE Transactions on Smart Grid, 2012.

[19] ——, “Profit maximization and power management of green data centers sup-porting multiple slas,” in Submitted to IEEE GLOBECOM Symposium on Green Networks and Communication Systems, CA, USA, December 2012.

[20] United States Environmental Protection Agency, “EPA report on server and data center energy efficiency,” Final Report to Congress, August 2007.

[21] A. Qureshi, R. Weber, H. Balakrishnan, J. Guttag, and B. Maggs, “Cutting the electric bill for internet-scale systems,” in Proc. of the ACM SIGCOMM, Barcelona, Spain, 2009.

[22] P. Torcellini, S. Pless, M. Deru, and D. Crawley, “Zero energy buildings: A critical look at the definition,” in Proc. of the ACEEE Summer Study on Energy Efficiency in Buildings, August Pacific Grove, CA.

[23] D. Kusic, J. O. Kephart, J. E. Hanson, N. Kandasamy, and G. Jiang, “Power and performance management of virtualized computing environments via lookahead control,” in International Conference on Autonomic Computing, June 2008, pp.

3 –12.

[24] H. S. Kim and N. B. Shroff, “Loss probability calculations and asymptotic anal-ysis for finite buffer multiplexers,” IEEE/ACM Transactions on Networking, vol. 9, no. 6, pp. 755 –768, December 2001.

[25] S. Boyd and L. Vandenberghe, Convex Optimization. New York, NY, USA:

Cambridge University Press, 2004.

[26] “Day ahead electricity price data archive,” Ameren, https://www2.ameren.com/RetailEnergy/realtimeprices.aspx.

[27] “The worldcup-98 internet traffic data archive,” Sited at Lawrence Berkeley Na-tional Laboratory, http://ita.ee.lbl.gov/html/contrib/WorldCup.html.

[28] X. Zheng and Y. Cai, “Reducing electricity and network cost for online service providers in geographically located internet data centers,” in IEEE/ACM Inter-national Conference on Green Computing and Communications, August 2011, pp. 166 –169.

[29] “Wind speed data archive,” alternative energy institute, http://www.windenergy.org/datasites/14-whitedeer/index.html.

[30] M. Abramowitz and I. A. Stegun, “Handbook of mathematical functions with formulas, graphs, and mathematical tables,” August 2007.

[31] Z. Liu, M. Lin, A. Wierman, S. H. Low, and L. H. Andrew, “Greening geo-graphical load balancing,” in Proc. of ACM SIGMETRICS joint international conference on Measurement and modeling of computer systems, New York, NY, USA, June 2011.

[32] S. Steinke, N. Grunwald, L. Wehmeyer, R. Banakar, M. Balakrishnan, and P. Marwedel, “Reducing energy consumption by dynamic copying of instruc-tions onto onchip memory,” in Proc. of the International Symposium on System Synthesis, Kyoto, Japan, October 2002.

[33] X. Zheng and Y. Cai, “Energy-aware load dispatching in geographically located internet data centers,” Sustainable Computing (to appear), 2011.

[34] N. Buchbinder, N. Jain, and I. Menache, “Online job-migration for reducing the electricity bill in the cloud,” in Proc. of International IFIP TC 6 Networking Conference, Valencia, Spain, May 2011.

[35] C. Cranor, M. Green, C. Kalmanek, D. Shur, S. Sibal, J. V. der Merwe, and C. Sreenan, “Enhanced streaming services in a content distribution network,”

IEEE Internet Computing, vol. 5, pp. 66–75, July 2001.

[36] S. Rahman, “Going green - the growth of renewable energy,” IEEE Power and Energy Magazine, vol. 1, pp. 16–18, November 2003.

[37] A. Ipakchi and F. Albuyeh, “Grid of the future,” IEEE Power and Energy Mag-azine, vol. 7, no. 2, pp. 52–62, March 2009.

[38] Z. Liu, M. Lin, A. Wierman, S. Low, and L. L. H.Andrew, “Geographical load balancing with renewables,” in Proc. of ACM GreenMetrics Workshop, San Jose, CA, April 2011.

[39] E. Muljadi and C. P. Butterfield, “Pitch-controlled variable-speed wind turbine generation,” IEEE Transactions on Industry Applications, vol. 37, no. 1, pp.

240–246, January 2001.

[40] M. Lange and U. Focken, Physical Approach to Short-Term Wind Power Predic-tion. Berlin, Germany: Springer-Verlag, 2005.

[41] C. Wu, H. Mohsenian-Rad, J. Huang, and A. Y. Wang, “Demand side man-agement for wind power integration in microgrid using dynamic potential game theory,” in Proc. of IEEE GLOBECOM Workshop on Smart Grid Communica-tions and Networking, Houston, TX, December 2011.

[42] H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load control with price prediction in real-time electricity pricing environments,” IEEE Transac-tions on Smart Grid, vol. 1, no. 2, pp. 120–133, September 2010.

[43] H. Allcott, “Real time pricing and electricity markets,” Working Paper, Harvard University, February 2009.