IBM Systems Director Active Energy Manager 2009년 10월12일 IBM Systems Director Active Energy Manager

Energy management: Measure, manage, optimize energy use Trending consumption on individual or group level Establish baseline cost Retrieve temperature and power information Spatial visualization “If you don’t measure it … you can’t mange it” Monitor and manages energy of every resource Correlate and control from a single console in the data center. So far we have spent much of this session looking at Improved Performance / Watt, Energy Saving Modes, Energy Monitoring & Trending Dynamic Energy Optimization, and Improving System Utilization. But a green agenda is not complete without looking at the overall Data Center / Building Efficiency & Reliability. How can we ties between IT and Data Center Infrastructure together. How can we integrate with Enterprise Management to do complete Data Center Efficiency , Thermal Monitoring & Mgmt and use new tools for Data Center Modeling. This chart represents the expanded capabilities IBM is now delivering through a integrated solution stack. IBM Director/AEM monitors and manages energy at the resource level for IBM systems and non IBM systems. Then Tivoli products expand the AEM scope and function to the IT Services, Workloads, and Service Level Agreements in the data center. Tivoli integrates Energy management into Enterprise Mgmt. This allows IBM to monitor power usage and thermal data from IT resources through embedded or remote sensors, leveraging partner capabilities for data center assets and facilities assets and integrate with application performance metrics. This all creates a method that integrates traditional IT measurements and emerging environmental measurements onto common dashboard with thresholding, trending, and event generation. This aggregation of IT and environmental metrics makes it possible to take manual or automated actions when needed for physical and virtual systems monitoring and management. Specific to facilities management, this all give IBM the unique ability to map and visualize data center facilities , obtain information on power, temperature, and layout, and identify problem areas , and enable improved facilities management in support of IT. Solution scenarios include Measure & Monitor at the lower levels of performance, utilization, response times, power usage and thermals. Control & Optimization then can be used for power capping, virtualization, storage tiering, and intelligent provisioning. As you move up Dynamic Optimization takes over by saving power by dynamic consolidation using Live VM Mobility or by coordinating with facilities infrastructure. What IBM has accomplished is the creation of Energy Management as a component of Systems Management. IT Assets 3rd Party Servers and Storage

Active Energy Manager의 동작원리 데이터 센터의 모든 IT 장비 및 Non-IT장비의 전력 사용을 네트워크를 통해 통합 모니터링 및 관리를 할 수가 있습니다. Software 설치에 의한 관리가 아닌 *BMC, MM, HMC, iPDU를 이용한 Hardware레벨의 관리이기 때문에 시스템에 미치는 부하가 극히 미비함. AEM Plug-In Director 서버 Out-of-Band access Network HMC HMC iPDU iPDU MM x86 cell Power BMC *Baseboard Management Controller는 IPMI와 같은 표준 인터페이스을 사용하여 시스템 온도,전압, 팬, 전원등에 대한 모니터링과 하드웨어 시스템에 대한 정보 수집을 수행하는 메인보드에 장착된 칩셋임 FSP System z Non-IBM 서버 Non-IT 장비 참조 System x Power6

Active Energy Manager 모니터링 기능 Power and Thermal Trending(무료) 지원 기능 개별 시스템의 전력사용량과 온도 변화 추이를 지속적으로 모니터링 기대 효과 데이터 센터의 전력사용량에 대한 효과적인 분석을 통해 사용량 예측과 같은 미래 계획에 반영 PDU+ (intelligent Power Distribution Units) 자체적으로 AEM을 지원하지 않는 Power 6 이전의 IBM 장비나 비IBM 장비에 대한 Power Monitoring을 가능하게 지원 데이터 센터 내에 산재한 다양한 플랫폼에 대한 에너지 관리를 통해 데이터 센터 전체의 전력 사용량 추이를 모니터링 Active Energy Manager Monitoring Features/Functions Active Energy Manager “no-charge” monitoring features/functions include: Power and Thermal Trending which enables a user to display power usage and thermal trending for individual systems over time (graph or table format. The also a client to Understand energy usage trends within and across systems helps DC managers calculate usage, scale back, negotiate better rates, or do future planning PDU+ (intelligent Power Distribution Units) - which enables support for power trending for older systems and storage devices and non-IBM systems through PDUs from IBM and non-IBM PDUs from Raritan and Eaton. support for power trending for older systems and storage devices and non-IBM systems hat are supported by the PDUs. Clients can now gathering energy management information across many platforms which enables datacenter managers to have a more complete view of power usage within the data center enhancing potential for even greater cost and energy savings Support for Facility Providers which enables the retrieval of temperature and power information via wireless sensors (SynapSense) and collect alerts and events from facility providers (Liebert SiteScan from Emerson Network Power) related to power and cooling equipment, Clients can provide a broader scope of energy information across systems within the datacenter while including a view of facilities equipment. Note -wireless sensors may be more feasible in the event PDUs are not a viable alternative. Additional information: PDU+ support enables energy monitoring of storage devices (DS3000, DS4000, DS6000, and nSeries storage control units) as well as System i, System x, and System p servers (Power 4 and Power 5 servers) when there is no native support available. What non-IBM systems are supported via the iPDU? AEM 4.1 provides “no-charge” monitoring functions for devices that can connect to select Power Distribution Units (PDU), specifically to an IBM DPI C13+ and IBM DPI C19 PDU+, and PDU+ models (feature codes 6050, 6051, 6090, and 6091) as well as Raritan and Eaton PDU+.. This enables monitoring functions for systems (IBM and non-IBM) that do not have native support for AEM. By plugging systems into a PDU (a smart power strip), AEM can collect energy information from any piece of equipment (including non-IBM servers) plugged into the PDU, including I/O drawers, providing a more complete view of energy used in the data center.

실제 전력 사용량을 측정 및 trend 데이터 생성 Measure and manage 실제 전력 사용량을 측정 및 trend 데이터 생성 서버 단위의 토출 온도 서버 예측치가 아닌 전력사용량에 대한 실제 통계치 Active Energy Manager 발열량 모니터 실제 사용량과 레이블 전력량의 비교 Measure and Manage Track things like temperature and power usage, with trending over time to help with planning. Power capping can be very important to clients who have signed an agreement with their Utility Providers where usage beyond a set threshold means they pay a premium. This capability allows the client to make the appropriate trade-offs between performance and efficiency. Note: power capping – set policies to throttle down the clock speed when certain thresholds are met to lower power consumption. Key Takeaway: AEM allows customer to trade off power for performance, providing the tools to make these decisions on how much energy each server should be allocated and setting thresholds not to exceed this level. Customers can work with actual data versus nameplate data on the server. 전력 및 온도 데이터 추이 데이터

Power Trending – PoC 결과

Power Trending IBM Active Energy Manager는 분단위의 원본데이터를 이용해 시/12시간/주/월/사용자정의 등 다양한 형태의 보고서를 출력할 수 있습니다. 시간 단위 12시간 단위 낮시간대에 업무증가량으로 인한 전력사용량 증가 그래프 시간범위가 1시간으로 설정됨 24시간 단위 주 단위 야간에 Batch 작업으로 인한 전력사용량 증가 그래프

Power Trending - 정확한 전력 사용량 정보 AEM 모니터링 기능을 이용한 실측 데이터를 바탕으로 하드웨어 제원 상의 소비전력으로부터 얻지 못한 전력 정보의 활용이 가능 시스템 제원 상의 전력 사용량 적정 전력량 할당 초과 할당 Power budget not converted into compute cycles Power budget converted into compute cycles Label Power Power Configurator 일반적인 하드웨어의 소비 전력을 기준으로 한 추정치 Improvement over label power Over Allocated 실측 데이터 Simple Example - Setting a Power cap. Within the chart there will be several clicks to emphasize that using AEM will provide the most accurate view of energy usage within a system 1 - A systems uses power over time and the old way of understanding it’s usage was by the label specification usage of the power shipped with the system which assumes you are running at 100% utilization - 100% of the time This results in over allocated power per system. 2 - Using power configurators available on multiple platforms can get you closer to the actual energy usage on systems but still not the actual usage over time which results in over allocation of power still. Furthermore power calculators can not adjust to changes in workload characteristics which might draw more or less energy 3- By using AEM monitoring functions one can trend the actual energy usage over time and then set a power cap so that power usage can be maximized thus saving on the overall power budget Note - Power trending will keep historical data on energy usage by a server. Based upon that information power capping will provide an effective range within which the system can execute while saving power but not affecting performance. The power cap can also be set manually by the account. Power (watts) Time   AEM 모니터링 

Simple Example - Setting a Power cap. HP Integrity 수퍼돔 정보 X 2.3 Simple Example - Setting a Power cap. Within the chart there will be several clicks to emphasize that using AEM will provide the most accurate view of energy usage within a system 1 - A systems uses power over time and the old way of understanding it’s usage was by the label specification usage of the power shipped with the system which assumes you are running at 100% utilization - 100% of the time This results in over allocated power per system. 2 - Using power configurators available on multiple platforms can get you closer to the actual energy usage on systems but still not the actual usage over time which results in over allocation of power still. Furthermore power calculators can not adjust to changes in workload characteristics which might draw more or less energy 3- By using AEM monitoring functions one can trend the actual energy usage over time and then set a power cap so that power usage can be maximized thus saving on the overall power budget Note - Power trending will keep historical data on energy usage by a server. Based upon that information power capping will provide an effective range within which the system can execute while saving power but not affecting performance. The power cap can also be set manually by the account. IDEAs International Performance. Heat and Power ratings RPE2 RPE OLTP BTU Watt IBM p6 570 – 16Cores 25,420 118,294.00 17,748 5,200 HP Superdome – 36 Cores 24,750 138,534.00 41,588 12,198

Active Energy Manager 관리 기능(유료) Power Capping 지원 기능 설정한 전력 상한 값을 초과하는 서버부하가 발생한 경우, 자동적으로 CPU Duty Cycle 신호를 조절하여 소비전력을 제어하는 기술. 기대 효과 센터의 온도가 높은 경우 전력 사용량에 대한 임계 값을 설정함으로써 시스템이 그 이상의 전력을 사용하지 못하도록 하여 저전력 운영을 통한 비용 절감효과를 기대 Power Saving 다양한 워크로드에 따라 전력 사용량에 대한 조절이 가능하게 함. 일반적인 CPU의 전력 사용량 대비 최대 30% 낮은 전력을 사용하게 설정함으로써 에너지 효율의 상승 효과 및 비용 절감을 기대. Active Energy Manager Management Features/Functions Active Energy Manager “priced” management features/functions include: Power Capping which allocates a maximum power level a system or group of systems can use and throttles the processor to use less power if the system starts to consume more than the maximum level set. Enhanced power capping or soft power capping which is an option within power capping, which enables a user to set a lower energy cap. The benefit to a client is that if it gets too warm in the data center, setting the cap will ensure that the system will not use more than that cap value, thus reducing power and thermal usage and costs Power Savings Mode which allows management of power usage within a system as work activity shifts based on varying demands across shift change. This can enable a system to save up to 30% of normal CPU power usage thereby improving energy efficiency and reducing costs Note - AEM V4.1 “priced” management functions can be downloaded via the AEM website for a 60 day no-charge evaluation period. 60일간의 trial 버전을 무료로 사용할 수 있으며 만료 후에도 모니터링 기능은 계속하여 사용 가능

Power Saver 구동 후 소비 전력 감소에 대한 검증 Power Saver 기능은 CPU 사용률이 낮은 시간대 CPU Clock Speed 조정을 통하여 실제 소비 전력을 절감 할 수 있습니다. 현재 사용한 장비는 IBM Power570 4.7Ghz 시스템으로 시스템 유휴상태(CPU Idle 100%)에서 saver mode의 소비전력 절감량을 측정하였습니다. 해당 시스템의 경우 소비전력이 약 10% 감소하였으며, 이는 시스템 구성에 따라 달라질 수 있습니다. 10% 전력 절감효과 PowerSaver OFF 2363W PowerSaver ON 2150W

Power Saver를 통한 전력 절감 및 성능에 미치는 영향 Measurement Target(CPU%) Throughput (Power Saver off) (Power Saver on) Throughput 영향도(%) 전력 절감율 Idle 18.8 10% 9,638 9,482 -1.62% 18.0 30% 20,885 20,904 +0.10% 19.6 40% 27,333 27,423 +0.33% 19.4 50% 33,268 34,093 +2.48% 18.3 60% 42,308 39,661 -6.26% 19.8 80% 55,341 54,800 -0.98% 19.2 90% 61,980 60,504 -2.38% 19.0 100% 66,685 60,684 -9.00% 20.6

S사 시스템의 테스트 결과 Power Saver 테스트 결과분석 적용 가능 구간 편차지표 시스템 사용률 10% 5% 0% 9% 증가 Power Saver off 상태 5% 1% 증가 4% 증가 편차지표 3% 증가 0% 8% 향상 5% 절감 4% 절감 4% 감소 5% 절감 5% 절감 2% 감소 -5% 10%부하 30%부하 50%부하 90%부하 -10% 시스템 사용률

Power Saver 기능을 통한 전력 절감 효과 - 예 560대 SUV가 부산왕복

Active Energy Manager 구축 사례 – H사 ERP 1단계 FHA 2 MMA 13 ERP 2단계 FHA 4 MMA 7 ERP 2단계(Dec.) MMA 6 IBM Systems Director Active Energy Manager IBM Confidential

환경을.. 그리고 미래를 위하여 IBM System Director Active Energy Manager를 통해 현재 여러분의 데이터 센터의 상황을 파악해 보십시오. 에너지를 절감할 수 있는 에너지 관리 계획을 수립해 보십시오. IBM System Director Active Energy Manager 의 관리 기능을 통해 에너지 절감 방법에 대해 생각해 보십시오. IBM System Director Active Energy Manager 를 통해 에너지 관리를 시작하면 비용 절감, 더 나아가 우리의 환경도 지킬 수 있습니다. IBM System Director Active Energy Manager 가 여러분의 녹색 성장에 기여할 것입니다. Next Steps Begin “monitoring” with IBM System Director Active Energy Manager 4.1 to understand what you have today - it’s FREE Create an energy management plan for your data center to reduce energy Implement energy conservation measures through Active Energy Manager’s “management” capabilities Go green with Active Energy Manager to manage energy usage and reduce costs while helping to preserve the environment

Building a Smarter Infrastructure – Green Infrastructure Agent 2CPU 4CPU Active Energy Manager IBM Systems Director Family Monitoring Analysis Control Agent 1CPU 3CPU 6CPU 친환경 저전력 서버 기술을 통한 전력절감 가상화기술을 통한 전력절감 통합 에너지 모니터링 및 관리 솔루션을 이용한 전력 절감