Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Energy and Thermal Management Metrics for Energy Efficiency in Data CentersC

989 views

Published on

Presentation at the Data Cloud Monaco 2015 on energy and thermal management metrics for energy efficiency in DC. Held by Marta Chinnici, from ENEA, and Alfonso Capozzoli, from Politecnico di Torino.

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Energy and Thermal Management Metrics for Energy Efficiency in Data CentersC

  1. 1. Energy and Thermal Management Metrics for Energy Efficiency in Data Centers DataCloud2015_Monaco 2-4 June Marta Chinnicia, Alfonso Capozzolib aENEA C.R. Casaccia, Via Anguillarese, 301, Rome – 00123, Italy bPolitecnico di Torino, Corso Duca degli Abruzzi, 24, Turin – 10129, Italy
  2. 2. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Data Center (DC) Energy Efficiency: Problem Statement Key Findings:  The role of DC within society is leading to a great increasing of interest both in ICT and Energy sectors: DCs are complex-system contain IT equipment used for the processing and storage data, and communications networking; Main components of electricity consumption for the ICT sector (Ref: Greenpeace Report, May 2015)
  3. 3. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Data Center (DC) Energy Efficiency: Problem Statement  The rapid rise of energy-hungry Data Center is driving significant growth of the power consumption, electricity usage.  The growing power demand of DCs has led to a heightened awareness of their increasing impact on climate change from greenhouse gas (GHG) emissions. Data Center energy use extrapoled to 2015 (ref: Koomey J.G.,2008 ) Growth in Data Center GHG emissions -2002 to 2020 (ref: GeSI report, 2012 ) However, DC emissions growth expected to slow down from 9% to 7%
  4. 4. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Why to measure Energy Efficiency in DC? The concept of Energy Efficiency became an important issue in DC design due to the increase of energy price and policy pressures. Energy Efficiency Policies Best Practices Green IT Efficiency Cooling Systems Optimizing Computing resources usage Metrics Renewable Energy Federation Smart Cities
  5. 5. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Measuring DC Energy Efficiency: Beyond Metrics The complexity of DC creates serious difficulties in pinpointing a methodology in terms of EE; within DC many variables need to be taken into account. In recent years, a variety of Metrics were proposed to evaluate DC energy efficiency by measuring performance at different levels and from different perspectives. Major barriers: There is a lack of a complete plan which provides standard metrics and methodologies for DCs. Key objectives:  The first step in energy efficiency improvement is to effectively evaluate energy consumption and DC environment by measuring the performance through a “HOLISTIC” approach.  Identify appropriate best practices and “optimization procedure” based on holistic approach for improving the design stage and management of a DC.  Identify standard ways for monitoring, measuring, verifying and reporting energy consumption and performance in DCs.
  6. 6. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Measuring DC Energy Efficiency: Beyond Metrics  Focus on the methodologies for capturing the energy consumption and carbon emissions arising from the DCs.  Introduce metrics capable to capture the complex phenomena occurring in a DC (e.g: the contribution of IT, cooling systems, “useful work”, thermal management and so on…).  To analyse the mutual relation among energy and thermal metrics.  To consider the contribution of renewable sources and hence, adapting the DC power consumption to the availability of renewable energy in dynamically way.  DCs in Smart Cities context: to adapt the requests received by the Smart City Energy Management authority.  Introduce metrics regarding DC Federation. Evaluation of DCs should be based on globally accepted assessment systems in terms of common metrics that promote the improvement of energy saving, renovation and improvement of infrastructures, management methods and so on…
  7. 7. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Current Energy Metrics: Criteria for selection and Methodologies In literature, several different metrics were proposed during the last few years however, there are a number of concerns with both a systematically approach and conceptual metrics tasks: 1. there is not a comprehensive classification of metrics; 2. a new perspective to connect thermal and energy consumption metrics. Energy Efficiency Energy Power Metrics Thermal Metrics DC Energy long term assessment Run-time thermal management diagnosis DC Design Stage DC Operati onal Stage Globally DC-EE Assessment
  8. 8. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics Overview Ref: Public Deliverable D7.1 of DC4Cities FP7-SMARTCITIES-2013(ICT), Description of energy efficiency metrics for Data Centers, August 2014.
  9. 9. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics Overview
  10. 10. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics Overview
  11. 11. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics Overview
  12. 12. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics Overview (IT level)
  13. 13. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics Overview (IT level)
  14. 14. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Current Energy Metrics: Criteria for selection and Methodologies Analysis of the current state of the art of metrics for DCs*.  Metrics were classified/discussed at different levels: 1. A first level of classification according to the nature of variables addressed:  power/energy metrics (related to energy consumption (kWh) or power demand (kW) at different levels);  IT useful work metrics (related to computing processes, data transfer, or storage);  thermal metrics (related to temperature);  waste and emission metrics (to measure the amount of natural sources wasted or the quantity of pollution generated by building and managing a DC). 2. A second level according to the physical infrastructure of DCs:  whole infrastructure system level;  component level (IT, HVAC, lighting,etc.). 3. A third level, according to selected objectives :  Minimize energy use, minimize emission/source consumption, renewable energy, energy reuse, scalability. *Ref: 1) Capozzoli A., Chinnici M., Perino M., Serale G.: Review on Performance Metrics for Energy Efficiency in Data Center: The Role of Thermal Management, Energy Efficient Data Center, Third International Workshop, E2DC 2014, Cambridge, UK. Ed. LNCS 8945, pp. 135-151, 2015. 2) Public Deliverable D7.1 of DC4Cities FP7-SMARTCITIES-2013(ICT), Description of energy efficiency metrics for Data Centers, August 2014.
  15. 15. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Criteria for selection and Methodologies Ref: Public Deliverable D7.1 of DC4Cities FP7-SMARTCITIES-2013(ICT), Description of energy efficiency metrics for Data Centers, August 2014.
  16. 16. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics: Progress Today & Challenges DC metrics in DC4Cities project: is developing its activities establishing a method to compare the measurements processes and assess new energy efficiency indicators for DCs, in order to outline a standardization procedure. DC4Cities project is leading the Smart City Cluster collaboration. The main objectives of the Cluster are: • ensure that all projects are able to use common KPIs (Key Performance Indicators) to characterize the energy, environmental and economic behaviour of their DCs; • enable the comparison between different projects; • collaborate with DC standardization organizations. New metrics focus on the energy behaviour of the DC:  Flexibility mechanisms in DCs:  Demand shifting: workloads are shifted from a time period to another, but always within the same Data Centre  Demand being federated: shifting the workloads to other Data Centres  Renewables integration: Energy produced locally and renewables usage  Primary energy savings and CO2 emissions avoided
  17. 17. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics: Progress Today & Challenges Challenge. The understanding and the actual possibility of calculating in mathematical way the "useful work" within a DC is particularly complicated: it depends on the applications/services within DC. What is the "useful work” (or Work Done) by a DC? Total amount of computing by all applications/services running in DC How calculate the "useful work" (or Work Done) done by DC? DCeP* (DC energy Productivity) To characterize the energy requested to produce useful computational work Total Work Done/Total Energy
  18. 18. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metrics: Progress Today & Challenges None of the current metrics directly gives a (mathematical) measure of the useful work in a DC. Issue: How it is calculated the value of V (normalization factor)? “Work Done” of different applications/services can’ t be simply added At present, there is no practical solution to these questions... “useful work” = describes the number of tasks executed by the DC and EDC represents the consumed energy respectively for the completion of the tasks. Practical solution: To «unpack» the sum of different tasks and to evaluate the normalization factor in order to compare the different workloads. Benchmark Procedure: evaluation of workload-based metrics
  19. 19. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli • It is possible to define mutual relations among performance metrics and reciprocal physical influences? • A thermal awareness approach can help to achieve the energy saving in DC? The key role of thermal management at design stage and during the operation of a DC in order to achieve energy saving will be discussed The role of thermal management • Which is the impact of thermal phenomena on the behavior of the global energy consumption and on the reliability of the IT equipment?
  20. 20. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Thermal metrics Typical design layout of server rooms AIR INLET to datacom equipment IS the important specification to meet OUTLET temperature is NOT important to equipment All power required to run IT equipment is dissipated as heat and because IT equipment needs to operate at appropriate temperature, designing an efficient cooling system becomes of crucial importance.
  21. 21. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Effect of temperature on component efficiency Infrastructure side An increasing of the DC environmental temperature causes a positive effect on cooling system. The efficiency of the cooling components rises. Tang, Q., Gupta, S. K. S., & Member, S. (n.d.). Energy-Efficient , Thermal-Aware Task Scheduling for Homogeneous , High Performance Computing Data Centers : A Cyber-Physical Approach, 1–14. Lee, K.-P., & Chen, H.-L. (2013). Analysis of energy saving potential of air-side free cooling for data centers in worldwide climate zones. Energy and Buildings, 64, 103–112. doi:10.1016/j.enbuild.2013.04.013
  22. 22. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli IT equipment side With the increasing of the DC room temperature, the consumption of the IT equipment rises. This is due to a major fan power consumption and server energy leakage.
  23. 23. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Effect of Temperature on server reliability
  24. 24. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Temperature variation effects on total energy consumption It is necessary to calculate an optimal temperature set-point. This would be the ideal tradeoff between cooling system and IT equipment energy consumption. This compromise could be reached by modifying the air inlet temperature in the cold aisle. It mainly depends on the server and CRAC characteristics. Durand-Estebe, B., Le Bot, C., Mancos, J. N., & Arquis, E. (2013). Data center optimization using PID regulation in CFD simulations. Energy and Buildings, 66, 154–164. The thermal management can affect the opportunity to set an optimal temperature?
  25. 25. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli IT Equipment Environment – ASHRAE Psychrometric Chart
  26. 26. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli The specifications are related to supply temperature in the datacom equipment and not to return CRAC/ambient temperature. ASHRAE Indications
  27. 27. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Airflow optimization and management This scenario wastes cooling capacity This scenario increases the inlet temperature to equipment HOT - COLD SPOT A fraction of inlet cold air does not contribute to cooling of the IT equipment The cold air intake into IT equipment is not sufficient and as a consequence a fraction of the hot air is recirculated
  28. 28. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Airflow within a DC
  29. 29. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metric Formula Information provided Supply Heat Index Warm air infiltration inside cold aisle Return Heat Index Return heat at CRAC units after recirculation Negative Pressure Ratio Warm air infiltration inside air-supply plenum due to negative pressure. It can be calculated from CFD analysis Bypass Ratio Mass flow rate that returns at CRAC units without heat power exchange Recirculation Ratio Mix between cold air supply and exhaust air from hot aisle Balance Balance between airflow at CRAC unit and across IT equipment Return Temperature Index Balance between airflow across IT equipment and at CRAC unit Global thermal metrics Based on DC average air temperatures
  30. 30. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Metric Formula Information provided β index Local increase of air temperatures along the rack Rack Cooling Index Low Rack cooling efficiency considering a lower threshold values Rack Cooling Index High Rack cooling efficiency considering an upper threshold values Capture Index (cold aisle) Cold airflow ingested by a rack. . It can be obtained from CFD analysis Capture Index (hot aisle) Warm airflow captured by a local extractor or cooler. It can be obtained from CFD analysis Local thermal metrics based on rack punctual air temperatures or mass flow rates
  31. 31. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli 1. RCIHi 2. RCILo 3. RTI 4. SHI/RHI Overall Airflow Efficiency Evaluation Standard Ideal Good Acceptable Poor Ideal Good Acceptable Poor Target Good Poor Good SHI<0.2 RHI>0.8 1 Poor - - - No Good2 - Poor - - 3 - - Poor - 4 Acceptable - - - Acceptable 5 Good/Ideal Poor Bypass or Recirculation - 6 Good/Ideal Good/Ideal Bypass or Recirculation - Good 7 Good/Ideal Good/Ideal Good - Very Good 8 Ideal Ideal Target Good Ideal Overall Evaluation Efficiency • To assure IT equipment reliability; • To assess absence of over cooling for energy saving; • To investigate presence of bypass or recirculation phenomena • Information about overall airflow efficiency
  32. 32. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Temperature variation effects on thermal metrics The temperature of the air supplied in the cold aisle is the most sensitive parameter for the environment temperature and air flow efficiency. Cho, J., Yang, J., & Park, W. (2014). Evaluation of air distribution system’s airflow performance for cooling energy savings in high-density data centers. Energy and Buildings, 68, 270–279. 0 0.2 0.4 0.6 0.8 1 13 15 17 19 21 SHI RHI Supply temperature Certain thermal metrics (SHI, RHI, RTI) are not sensitive to the variation of a single temperature parameter, because based only on temperature differences. Other metrics, such as RCI, are very sensitive to temperature variations.
  33. 33. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Mutual Relation Among Thermal and Energy Consumption Metrics Some authors recognizes that the mixing of hot and cold streams in the DC airspace is an irreversible process and must therefore lead to a loss of exergy. The proposed exergy- based approach can provide a foundation upon which the DC cooling system can be simultaneously evaluated for thermal manageability and energy efficiency. • Vice versa, thermal metrics are still of limited use because few information is gained regarding the energy efficiency of the system. • In general power/energy metrics provide no information about bypass/recirculation phenomena and corresponding impacts on the thermal manageability of DCs. • Thermal metrics are used to enable real-time feedback and control of DC thermal architecture, while power/energy metrics to outline the global energy consumption. Usually these metrics are used in parallel
  34. 34. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Second-law analysis that considers both global thermal management and local phenomena (recirculation) as irreversible processes source of exergy losses. Exergy losses occur in: - Air space; - Rack units; - CRAC units. An exergy-based approach for the evaluation of thermal manageability and energy efficiency A.J. Shah, V.P. Carey, C.E. Bash, C.D. Patel, Exergy Analysis of Data Center Thermal Management Systems, J. Heat Transfer. 130 (2008) 021401. Comparison with other metrics for thermal management and hot-spot recognition Exergy losses balance into a DC The airspaces losses are divided themselves into: - Airspace into hot and cold aisles; - Airspace into rack units; - Airspace into CRAC units.
  35. 35. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli . The sum of the different exergy losses provides an appropriate metric for the overall DC. The share of recoverable exergy represents a possible improvement for the DC. Exergy losses distribution in the air-space of a DC Temperature distribution in the air-space of a DC A map of the data center airspace that highlights locations of recirculation. This maps is related to the temperature distribution The exergy-based metric is more sensitive to recirculation than the traditional temperature- based metrics. An exergy-based approach for the evaluation of thermal manageability and energy efficiency
  36. 36. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Mutual Relation Among Thermal Management and Energy Consumption • In order to offset by pass and recirculation air issues, the CRAC units are often designed and controlled to supply air at a lower temperature, thus with an higher energy consumption. • Global energy indices are not necessary capable to detect these phenomena. Indeed, hotspots are local phenomena whose influence on the global power/energy efficiency may be negligible. • The energy metrics are referred to long term period (e.g. a year or a season) while hotspots are phenomena which depends on short term variation of boundary conditions • Therefore it is necessary to apply a continuous commissioning to detect the occurrence of local phenomena through thermal and energy metrics.
  37. 37. Energy Efficiency Management Through Thermal Performance Awareness  Tang, Q., Gupta, S. K. S.,Varsamopoulos, G.: Thermal-Aware Task Scheduling for Data Centers Through Minimizing Heat Recirculation. 2007 IEEE International Conference on Cluster Computing, 129–138. doi:10.1109/CLUSTR.2007.4629225 (2007)
  38. 38. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Energy Efficiency Management Through Thermal Performance Awareness This way of task scheduling to IT equipment with thermal awareness guarantees both the minimization of hot air recirculation (low values of SHI) and energy consumption, also in relation with other algorithms of task assignment.
  39. 39. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Conclusions  Evaluation of EE in DCs should be based on globally accepted assessment systems in terms of common metrics and methodologies.  A holistic framework may help to take into account the effects on all metrics simultaneously, but the need for the development of new, more accurate and/or usable metrics is recognized  Alongside to existing metrics, is necessary to provide renewable energy metrics for DC in order to facilitate with renewable power their operations.  Identify DC energy metrics for meeting future energy needs in smart cities context.  Long term energy assessment and “real time” thermal environment diagnosis should be not considered as separate tasks for a comprehensive DC performance analysis. These two aspects should be always coupled.  The thermal management should be achieved through the calculation of local thermal metrics primarily, and then by other average thermal metrics referred to the whole DC environment.  On the other hand the energy assessment should be performed through power/energy metrics capable to capture in a correct way the effect of energy consumption variation for both cooling and computing as well as the adaptability at part load conditions of DC infrastracture  The improvement and optimisation of DC performance through a thermal awareness approach to minimize recirculation effect represents an effective way to obtain energy savings.
  40. 40. Energy and Thermal Management Metrics for Energy Efficiency in Data Centres Marta Chinnici Alfonso Capozzoli Thanks you for your attention! Any questions? MARTA CHINNICI marta.chinnici@enea.it ALFONSO CAPOZZOLI alfonso.capozzoli@polito.it

×