Data Center Energy Efficiency ScaleUpScaleOut Processor Design Background

Data Center Energy Efficiency: Scale-Up/Scale-Out Processor Design Background & Analysis By Nick

Overview Data Centers Energy Usage Minimize costs Design Philosophies Scale-up Scale-out Comparisons

Data Centers Originally in giant enterprises Ran operations across buildings in business Now available for rental by the common man Webhosting Small business rentals Cloud Computing

Energy Usage It costs *a lot* to run a data center Comparable to a small town’s energy usage Can’t use constant power Low power states Dynamic power management Traffic Engineering Architecture designs…

Scale-Up Processors Improve control units to make more efficient Upgrade/replace current machines with better versions Tile Processors Array of processing units Networked together Each computer becomes a “cluster”

Scale-Out Processors Add more units to existing system Provides more workers Less bottlenecks at top Closer to the ideas of parallelization Increased energy costs (more units) Offset by increased throughput and energy efficiency

Analysis Scale-out had higher performance efficiency compared to scale-up High workload demand environments However, too many units led to decrease in efficiency Possibly due to “too many cooks in the kitchen” Draining power without significant time decrease Overhead constraints increase

Conclusion Scale-out architectures seem to scale better as demand increases However, there seems to be a limit with a constant workload In those cases, scale-up provides further gains Hybrid approach may be the best bet

References [1] https: //www. gartner. com/it-glossary/data-center/ [2] Greenberg, A. et al. The Cost of a Cloud: Research Problems in Data Center Networks [3] https: //www. gartner. com/it-glossary/total-cost-of-ownership-tco [4] http: //plus. usgbc. org/efficient-data/ [5] Buyya, R. , Jin, H. , Cortes, T. Cluster Computing. Future Generations Computer Systems. Vol 18. Pages 5 – 8. 2002. [6] Taylor, M. Tiled Microprocessors. Massachusetts Institute of Technology. February 2007. [7] Lotfi-Kamran, P. et al. Scale-Out Processors. Proceedings of the 39 th Annual International Symposium on Computer Architecture. Pages 1 – 12. June 2012 [8] Niewiadomska-Szynkiewicz, E. , et al. Dynamic power management in energy-aware computer networks and data intensive computing systems. Future Generation Computer Systems. Vol 37. Pages 284 -296. July 2014. [9] Staff. The power of power-aware scheduling. Argonne National Library. January 28, 2014. [10] Wenish, T. F. , Buyuktosunoglu, A. Energy-Aware Computing. IEEE Computer Society, Page 6 - 8, September/October 2012. [11] Dean, J. and Ghemawat, S. Map. Reduce: Simplified Data Processing on Large Clusters. Google, Inc. [12] Banerjee, A. et al. Cooling-Aware and Thermal-Aware Workload Placement for Green HPC Data Centers. IMPACT Laboratory. Arizona State University. [13] https: //smartech. gatech. edu/handle/1853/6560 [14] Cunniffe, R. and Coghlan, B. Encouraging the Unexpected: Cluster Management for OS and Systems Research. Department of Computer Science. Trinity College. Dublin, Ireland. [15] https: //aws. amazon. com/what-is-cloud-computing/ [16] Wang, Q. , Shen, L. and Wang, Z. Research on Scale-Out Workloads and Optimal Design of Multicore Processors. Proceedings of International Conference on Soft Computing Techniques and Engineering Application. Pages 157 - 166. December 21, 2013.