Comprehensive Energy Efficiency of Data Centers Case

Comprehensive Energy Efficiency of Data Centers – Case study shared. Anand Vanchi- Intel IT Ravi Giri – Intel IT Sujith Kannan – Intel Corporate Services

Legal Notices This presentation is for informational purposes only. INTEL MAKES NO WARRANTIES EXPRESSED OR IMPLIED IN THIS SUMMARY. *Other names and brands may be claimed as property of others Copyright © 2009, Intel Corporation. All rights reserved. 2 3/17/2018

Need for energy efficiency in Datacenters Continuously increasing Energy consumption for growing computing needs!!!!!! Energy use of computing devices is an increasing concern – Total U. S use is about 100 B Kwh. in 2006* – Projected to grow 3 -4% per year* – Supply – Demand gap for India continues to increase – Energy costs likely to continue increasing – Supply issues; potential climate change regulations 3 3 The same is reflected in the Datacenter energy consumption also. . U. S. data center electricity use ~30 B Kwh/yr. * Energy is top data center operating cost Public perception that data centers are driving up energy use Worldwide server shipments 7 M/ year and growing 10% per year * * Source – Energy Efficient computing – Tim Higgins, Intel Dt: 5/8/07 3/17/2018

Challenges to Energy Efficiency in Data centers – Right Sizing of facility infrastructure considering appropriate load diversity. – Facility side redundancies not in line with business needs. – Legacy servers yielding less performance per watt of power consumed as compared to the current generation multi core servers. – Lack of energy efficiency as a key decision vector for appropriate , optimum work load allocation to Servers. – Lack of correlation of Facility utilization and IT performance measurements. “ A clear consistent way of measuring DC Efficiency which includes business value, IT performance and Facility Utilization aspects would helps us address the above challenges ” 4 4 3/17/2018

Where do we start…… Datacenter Facilities Metering and tracking of PUE. – Challenges – Isolating Datacenter Power and cooling loads from the rest of the building facilities. – Metering of Facility utilization (Power /Cooling/Losses) at right points in distribution providing useful actionable information for efficiency improvement. – Identification of optimum levels of granularity for utility metering of the Datacenter. – Analysis, Archival of the metered will share and actionable Going forward we information a report generation study on how we are case trying to overcome the above mentioned challenges and the benefits we realized in our endeavor. . 5 3/17/2018

Intel DC Case Study DC Power metering – Input side – UPS Input panel Output side – Row Level Power distribution panel DC Cooling metering – PAC Units – at Distribution panel supplying power to the PAC Units. Multi use Chiller plant – Used Flow Meters and Temp sensors to isolate cooling consumption on account of the DC alone. 6 3/17/2018

Case Study contd. - UPS Power Metering Layout Energy Meters at UPS input panels Energy Meters at Row level Output panels nearer to IT loads 7 3/17/2018 Intel Confidential

Case Study contd. – Cooling Power Metering Layout Energy Meters in PAC Electric panels 8 3/17/2018 Intel Confidential

Intel DC Case Study - DC Flow meter Layout Chilled water supply header Chilled water Return header To PAC’s Supply header temperature sensor Return header temperature sensor 9 Flowmeter To PAC’s

Intel DC Case Study contd. - Energy Meter readings from -BMS 10 3/17/2018

Intel DC Case Study Contd. - Flow Meters Readings - BMS TEMP SENSORS -1 11 3/17/2018 TEMP SENSORS -2

The measurements collected. . Facility Power Consumption measurements – 1. UPS & Distribution Losses. 2. 3. 4. 4. 5. 6. IT Load Power consumption Total Power consumption by PAC Units in DC. Power consumption for cooling of UPS room. Power consumption at Chiller plant w. r. t DC Lighting load measurements. Power consumption on account of Make up AHUs. Utility Report Generation – 1. Continuous PUE measurements 2. Impact on PUE due to variation in IT load. 3. Impact on PUE due to variation in Facility corresponding to variation in load 12 3/17/2018

The outcome of facility utilization measurements. . The actionable triggered from the analysis of the measurements. . 1. Paralleling of UPS to increase the utilization levels thereby reducing efficiency and distribution Losses. 2. Increasing the PAC tempt Set point from 19 Deg C to 23 Deg C. 3. Humidity controls made operational as needed in DC (was found to be required only for 3 Months in Bangalore) 4. Effective Air management inside the DC. 5. LED lighting for emergency lighting inside the DC. 6. Regular lighting utilization also controlled and managed. The above activities were rolled out in the DC at various times during the year based on the inputs identified from metering and measuring of the facility consumption in the DC and this resulted in improving the PUE to 1. 81 from 1. 99 13 3/17/2018

Power Cost savings by improvement in PUE / Efficiency Achieved Annual Savings in 2008 by improving PUE from 1. 99 to 1. 81 14 3/17/2018 Potential annual power cost savings if PUE further improved to 1. 68

Benefits achieved after the Facility Efficiency measurement program – How did this effort help? – Annualized Operational power cost savings (77 K in 2008) – Establishment of the relationship between Power density optimization and improved efficiency. – Contributed to ROI justification for retrofitting the DC to higher density to refresh 4 yr old single core servers with High Performance Multi core servers in the DC. – Proliferating the knowledge gained as a global effort within the organization and contributing in terms of Data and Architecture inputs for the effort. 15 3/17/2018

What next after Facility Efficiency…. Why facility efficiency is just the beginning and means but not the end? ? ? ? ? – Refer the cost savings in our case study. . – An annual power cost savings of 77 K USD realized in 2008 by improving PUE from 1. 99 to 1. 81 – Potential annual power cost savings of additional 55 K USD targeted in 2009 by improving PUE from 1. 81 to 1. 68. – However improving PUE is still a finite activity. Beyond a reasonable value improvements in PUE will not yield an effective ROI as the returns are of diminishing nature. To ensure a significant ROI in the DC efficiency initiatives the need is to focus holistically including the IT utilization aspects and its impact on facility efficiency. 16 3/17/2018

ROI Impact of IT optimization on DC Efficiency Facilities optimization DCE = (1/PUE) 17 3/17/2018 IT Infrastructure optimization

The Progress into next level. . Datacenter IT Infrastructure asset tracking and IT performance measurement – Where we are working on currently? – Creation of a Database with all DC IT Infrastructure information. – Establish the relationships between IT and Facility equipments (Power / Cooling / Weight / Network distribution) – Tracking and metering of IT equipment performance at an optimum granular level (Eg. Spec. INT / Org Specific etc. ) – Analysis, Archival of the metered information and actionable report generation – Technology for implementing Power management policies based on actionable reports. 18 3/17/2018

Sample Power usage per ROW/RACK level 19 3/17/2018

Sample Data collection format - DC IT max power load Actual versus expected Sample Data collection format – Network Ports Usage Tracking at Row Level 20

Sample IT Performance Measurements Screen Overall servers performance at ROW level 21 3/17/2018

The matured model…. UTOPIA Where we want to be!!!!!!! Dynamically manage the efficiency of the Datacenter holistically by managing an effective and optimum IT performance to Facility utilization levels – Identifying the optimum threshold, the sweet point for the Infrastructure between efficient utilization threshold and Reliable utilization threshold – The ability of the IT Application system to ramped up or ramped down as per business needs and correspondingly respond in facility provisioning dynamically. - Automated policy based decision making to chose the best location for moving the IT Work load in a geographically distributed data center architecture. Establish clear Running Cost in USD per unit IT Performance and effectively evolve the TCO model in delivering the data center service in the most efficient way both from facilities and IT perspective. 22 3/17/2018

Questions? ? ? 23 3/17/2018