Virtualización para la Eficiencia
- 1. VIRTUALIZATION for EFFICIENCYKathrin WinklerDirector, The Green GridVP,EMC Corporation
- 2. About The Green GridA not-for-profit global consortium focused on driving energy efficiency in the computing ecosystemDeveloping meaningful and user-centric metrics to help IT and Facilities better manage their computing resourcesDeveloping and promoting standards, measurement methods, best practices and technologies that support these metrics22
- 3. x86 Servers are Driving Power ConsumptionCourtesy of VMwareServers = 40% of total data center electricity useVolume servers = 85% of all server electricity useUp from 70% in 2000 at 17% CAGR (compound annual growth rate)Servers drive other IT and site infrastructure consumption Start by solving the problem “upstream” at the source!3
- 4. Processor Capacity TrendsUse 5%Courtesy of VMwareMHz Capacity Requirements +10% YOY4
- 5. Power Consumption vs UtilizationCourtesy of VMware
- 6. Sample with 500 x86 Servers6
- 7. Determine the Opportunityfor Consolidation of Servers Using VirtualizationDetermine the Opportunityfor Consolidation of Servers Using Virtualization7
- 8. Gathering IntelligenceCollect information on all serversAgent-less tools are available Discover, inventory, and collect performanceGroup servers and make consolidation recommendationsCharacterize workloadsResource consumption over timeWorkload volatilityPotential risk8
- 9. Minimum InformationConfiguration & UtilizationMake, modelProcessor speed, core count, etc.MemoryNetworkLocal storageOS and patch levelSoftwareRisk ItemsOwnershipService level requirementsSecurity, privacy, regulatory or political issues9
- 10. % Processor Time (not normalized)UtilizationSingle 1.26 GHz(2) 2.8 GHzHourCourtesy of VMware10
- 11. Normalized Processor MHz Load(2) 2.8 GHzSingle 1.26 GHzTimeCourtesy of VMware11
- 12. Load CharacteristicsCourtesy of VMware12
- 13. Determine Workloads that Can CoexistExclude workloads which must reside on their own serversIdentify workloads with above average or volatile workloadsGroup workloads that can co-existSame LocationSame Service LevelSame Environment (Innovation, Production, Mission Critical)13
- 14. Implementation StrategyPhase 1Low resource requirementsLow volatility in resource consumptionLow riskPhase 2Low utilization & volatility of resourcesMedium riskPhase 3Everything else that can be virtualized14
- 15. Consolidation HurdlesAwarenessInfrastructure knowledgeVirtualization knowledgeTechnicalData center preparationApplication restrictionsFinancialServers not fully depreciatedBudget freezePoliticalPerceived loss of control15
- 16. Strategies to Clear the HurdlesConduct a Complete AssessmentInclude all servers, all locationsDevelop Data Center ModelBased upon post consolidationInclude growth estimatesDevelop Transition PlanFull consolidation takes more than a year in most casesInfrastructure needs will step down with server reductionMarket InternallyDevelop RewardsMore independenceLower cost16
- 17. Virtualization – Two StepsWP #19 USING VIRTUALIZATION TO IMPROVE DATA CENTER EFFICIENCYData CenterData CenterData CenterPOWERCOOLINGPOWERCOOLINGServersPOWERCOOLINGStorageServersServersStorageStorageNetworkNetworkNetwork#27 Impact of Virtualization on Data Center Physical Infrastructure17
- 18. Physical InfrastructureLoss CharacteristicsLosses that occur at higher loads (ex. switching, conduction)Powered on withoutIT Load18
- 19. Data Center Power Efficiency Metrics: PUE and DCiE19Total Facility PowerIT Equipment PowerPUE = Power Usage Effectiveness =Data Center Infrastructure Efficiency 1PUEIT Equipment PowerTotal Facility PowerDCiE = = =
- 20. Efficiency CurveReducing IT LoadRight-Size Power-Cooling Equip(not considering redundancy)20
- 21. Which Density Strategy?Low-density racks with no virtualized serversNew or existing data center?
- 22. How many high density servers?
- 23. How much bulk power and cooling?
- 24. Can IT devices be moved around?Low-density racks with a mix of virtualized and non-virtualized serversHigh-density racks with virtualized serversConsolidate the loadSpread the loadSupplemental CoolingSpectrum of post-virtualization densities21
- 25. Spread the Load StrategyEasy…
- 26. No new power and cooling equipment required
- 27. But…
- 28. Takes up valuable floor space - lots of white space
- 29. No power/cooling efficiency benefit
- 30. New configuration could compromise existing cooling strategy/architecture
- 31. Host servers inherit same availability as less critical servers22
- 32. When to Spread the LoadPost-consolidation average kW/rack < 3 kW
- 33. Placement of IT equipment is flexible
- 34. IT equipment power and cooling redundancy requirements could limit mobility of IT equipment
- 35. Additional power drops available only in certain locations
- 36. Cooling redundancy may be based on physical location
- 37. Open U-space is available in existing racks23
- 38. Supplemental Cooling StrategyLow Capital Expense
- 39. But…
- 40. Effective only for the lower range of high density
- 41. Temporary solution for cooling of hot spots
- 43. Installation of fans to boost the cool air supply from the CRAC to an enclosure
- 44. Variable speed drives (VFD) on CRAC units
- 45. Better hot isle containment such as special racks
- 46. Installation of row-based air conditioners to address localized hot spots
- 47. A rear-door heat exchanger shortens the air path to back of rack from CRAC24
- 48. When to Deploy Supplemental CoolingPost-consolidation average kW/rack is 3 kW – 5 kWNo flexibility over the placement of IT equipmentThere is limited capital budgetOptions:Specialty (grate-type) floor tiles or fansPortable spot air conditioning unitsSpecial racks or rack mounted cooling devicesRow-based air conditioners between racks25
- 49. Consolidate to High and Low Density Areas StrategyTakes more planning…
- 51. Perhaps retrofit only high density area
- 52. But has big advantages…
- 55. Better utilization of floor space
- 56. Enables targeted availability/redundancyLocal cooling for hi-powered, dense equipment26
- 57. When to Designate High & LowDensity AreasPost-consolidation average kW/rack >5 kWSubsection of data center is available for pod(s)Desire to improve overall data center efficiency (PUE)Use high density area to focus on cooling efficiency :Improve hot air containmentReduce air path length, e.g., top of rack cooling, row cooling, or rear door heat exchanger27
- 58. Characteristics of Dynamic LoadsRow based cooling, CRAC VFD are methods to adjust to variable IT loadsCaution: Loads constantly shift as data center experiences higher/lower activity
- 59. IT movement (e.g., vMotion) is typically unaware of cooling limits in any physical space28
- 60. Other Considerations When Moving to High DensityTypically complicates cable managementLeftover cables when moving from low to high densityAdditional cables because of additional IT equipmentRequires higher rates of air flow to support heat loadIncreased weight of racks due to additional IT in rackImpact on power infrastructure (breakers, panels, etc.)29
- 61. SummarySurveying “Things To Consider” before /during / after IT VirtualizationThe virtualization savings entitlement can be multiplied if power and cooling infrastructure is modified or upgraded Spreading the load, supplemental cooling, and designation of high density and low density areas can help further enhance data center efficiency performance No one size fits all30
- 62. For More Informationwww.thegreengrid.orgfor whitepapers, training,and contentFor membership, please contact us: +1 503-619-0653 or [email protected] 31Thank you to our members for your continued dedication and involvement in ICT energy efficiency efforts!
- 63. Thank you!For more information please go to www.thegreengrid.org