HP 22.1kVA HP intelligent power infrastructure solutions - Page 2

Introduction In the past, you have probably based expected power consumption levels on error-prone, timeconsuming manual estimates that often result in overestimating or underestimating resources. Energy efficient ProLiant and BladeSystem servers include technology that can measure and automatically adjust power consumption according to actual server demand. HP Intelligent Power Discovery removes the guesswork in determining a data center's power needs, which lets you reclaim trapped power and avoid unscheduled downtime. It gathers server configuration and status data and maps the data to define the complete power and IT equipment infrastructure-the topology of the data center. HP Intelligent Power Discovery is also scalable for future expansion and increased functionality, ensuring a return on your investment. This technology brief describes automatic power monitoring and control at the component level for efficient and reliable data center operation. Power distribution considerations in high-density systems Facility power distribution was simply designed for a worst-case power load, either using estimates from the nameplate ratings of individual components or observing maximum load values of existing equipment. Using these methods, you subtract the estimated load from the infrastructure's de-rated capacity to find the amount of headroom for a power circuit. For example, if the manufacturer derated a 1 MW power distribution circuit to 800 kW, an IT manager estimating a 700 kW maximum load could assume the infrastructure would allow100 kW of overhead for expansion. Typical server power consumption varied very little from idle to peak load; server idle power consumption was 90 to 95% of peak load. But now, server vendors have changed server behavior by giving you designs that scale power consumption with compute load. With greener chip sets, servers transitioning from idle to peak can dramatically increase power consumption. For Figure 1. Reduced energy use with G6 HP servers instance, Figure 1 shows us that the load required for a Generation 6 (G6) server running at 20% utilization can nearly double at 100% utilization. Because of this, estimating worst-case load from measured load has become a poor substitute for actual worst-case load. Data centers using power monitoring tools tend to underestimate energy needs since only an observed peak (and not the potential load) can be determined. Going back to the earlier 1 MW example, if an observed load of 700 kW is measured for a room full of G6 servers running at 20 percent utilization, a sudden rise in compute load could easily push the power requirements beyond 1 MW, creating an obvious overload problem that could be catastrophic. Data centers using nameplate ratings will have a fixed power allocation for each distribution circuit (Figure 2). Expanding these circuits is not possible. The racks may have space for more equipment, 2