The function of data centers within our modern society, long considered as black box utilities to house and manage data pertaining to various organisations, is moving to a more active role. The future outlook involves seamless integration within the smart grid and smart cities in order to enable new energy analytics and efficiency paradigms, as investigated within the GEYSER EU FP7 project.
Whether a data center initiates load migration to process IT load at the lowest cost or a market operator triggers a data center to provide electricity to the grid, decision-making will require measurement of the data center’s internal energy context. Traditional point-in-time and low-frequency sampling approaches to power monitoring have proven to be limited for optimised control over migration processes and demand response, as these cannot rely on ‘past’ data.
Data Center Energy Monitoring Software
More sophisticated real-time and minute-level monitoring are key to provide a comprehensive view of the electricity spent within a data center infrastructure, as they allow fine-grained understanding of how the data center is currently performing and how it is expected to behave in the near future based on historical data. Load demand from racks and cooling are closely monitored, and signs of failure or inefficiency can be discovered in time providing more control over the management of facility resources. In addition to this, analysing equipment high-resolution load patterns can uncover opportunities for fine-tuning equipment settings and improving the overall efficiency.
Data Center Power Monitoring Anomalies
Rapid changes in computing load translate to increased computational power, which in turn generates additional heat dispersion and higher cooling requirements.
Electrical meters provide measurements of consumption for most electrical components of a data center; with the right data sets, intelligent software can go one step further and uncover what physical measurements won’t tell you. Load demand periodicity, deviation from operational baseline and expected behaviour, signs of equipment failure, accurate Power Usage Effectiveness (PUE) variations and other key data center performance metrics are examples of insights that will be invaluable when deciding to balance processes and energy.
Complex real-time data analysis requires high data sampling from the meters, so that models of the load demand can be accurately constructed. Under constrained computing requirements, variations of the load demand from IT and facility infrastructure will trigger exceptions from the models that will translate into warnings, alerts, discovery of offenders, and much more. Automating this process makes sure that the data center internal energy context is known and monitored at all times, enabling confident execution of optimisation opportunities.
With adequate measurements, the periodicity of events monitored can be unveiled and used to facilitate decision-making against expected and predicted behavior. For example, the change of frequency of PUE variations will lead to reconsideration of the cooling elements efficiency, as they could indicate defects or non-effectiveness.
Identifying recurring patterns and correlations
Another category of data analysis enabled by high-sampling metering is that of non-intrusive load monitoring (NILM). NILM systems are designed to monitor the current and voltage waveforms of an electrical measurement to estimate, in software, the nature and energy consumption of electrical components switching on and off on that specific circuit. In data centers environments, where all servers have similar features, correlating real-time variations of the servers’ overall load demand with measurements of the servers’ individual computing power levels provides essential information for software load monitoring. Deploying non-intrusive NILM systems can as a result dramatically reduce the number of measurement points, and more importantly provide soft measurements for equipment that would otherwise not be monitored.
Intelligent integration of networked urban data centers with energy infrastructures, i.e. smart power grids and district heating, requires flexible management of the IT workload in the context of dynamic energy availability. Real-time minute-level power measurements will provide a suitable internal energy context for data centers to allow balanced energy exchanges within a physical surrounding and workload exchanges with other data centers, and promises informed optimisation and improved control over operations.
See Also: Retail Energy Management Software
See Also: Hotel Energy Management Software
See Also: Data Center Energy Management
See Also: Building Energy Management